1 USING THE IJG JPEG LIBRARY
3 This file was part of the Independent JPEG Group's software:
4 Copyright (C) 1994-2011, Thomas G. Lane, Guido Vollbeding.
5 libjpeg-turbo Modifications:
6 Copyright (C) 2010, 2014, D. R. Commander.
7 For conditions of distribution and use, see the accompanying README file.
10 This file describes how to use the IJG JPEG library within an application
11 program. Read it if you want to write a program that uses the library.
13 The file example.c provides heavily commented skeleton code for calling the
14 JPEG library. Also see jpeglib.h (the include file to be used by application
15 programs) for full details about data structures and function parameter lists.
16 The library source code, of course, is the ultimate reference.
18 Note that there have been *major* changes from the application interface
19 presented by IJG version 4 and earlier versions. The old design had several
20 inherent limitations, and it had accumulated a lot of cruft as we added
21 features while trying to minimize application-interface changes. We have
22 sacrificed backward compatibility in the version 5 rewrite, but we think the
23 improvements justify this.
30 Functions provided by the library
31 Outline of typical usage
36 Mechanics of usage: include files, linking, etc
38 Compression parameter selection
39 Decompression parameter selection
42 Compressed data handling (source and destination managers)
44 Progressive JPEG support
46 Abbreviated datastreams and multiple images
48 Raw (downsampled) image data
49 Really raw data: DCT coefficients
53 Library compile-time options
54 Portability considerations
56 You should read at least the overview and basic usage sections before trying
57 to program with the library. The sections on advanced features can be read
58 if and when you need them.
64 Functions provided by the library
65 ---------------------------------
67 The IJG JPEG library provides C code to read and write JPEG-compressed image
68 files. The surrounding application program receives or supplies image data a
69 scanline at a time, using a straightforward uncompressed image format. All
70 details of color conversion and other preprocessing/postprocessing can be
71 handled by the library.
73 The library includes a substantial amount of code that is not covered by the
74 JPEG standard but is necessary for typical applications of JPEG. These
75 functions preprocess the image before JPEG compression or postprocess it after
76 decompression. They include colorspace conversion, downsampling/upsampling,
77 and color quantization. The application indirectly selects use of this code
78 by specifying the format in which it wishes to supply or receive image data.
79 For example, if colormapped output is requested, then the decompression
80 library automatically invokes color quantization.
82 A wide range of quality vs. speed tradeoffs are possible in JPEG processing,
83 and even more so in decompression postprocessing. The decompression library
84 provides multiple implementations that cover most of the useful tradeoffs,
85 ranging from very-high-quality down to fast-preview operation. On the
86 compression side we have generally not provided low-quality choices, since
87 compression is normally less time-critical. It should be understood that the
88 low-quality modes may not meet the JPEG standard's accuracy requirements;
89 nonetheless, they are useful for viewers.
91 A word about functions *not* provided by the library. We handle a subset of
92 the ISO JPEG standard; most baseline, extended-sequential, and progressive
93 JPEG processes are supported. (Our subset includes all features now in common
94 use.) Unsupported ISO options include:
95 * Hierarchical storage
98 * Nonintegral subsampling ratios
99 We support both 8- and 12-bit data precision, but this is a compile-time
100 choice rather than a run-time choice; hence it is difficult to use both
101 precisions in a single application.
103 By itself, the library handles only interchange JPEG datastreams --- in
104 particular the widely used JFIF file format. The library can be used by
105 surrounding code to process interchange or abbreviated JPEG datastreams that
106 are embedded in more complex file formats. (For example, this library is
107 used by the free LIBTIFF library to support JPEG compression in TIFF.)
110 Outline of typical usage
111 ------------------------
113 The rough outline of a JPEG compression operation is:
115 Allocate and initialize a JPEG compression object
116 Specify the destination for the compressed data (eg, a file)
117 Set parameters for compression, including image size & colorspace
118 jpeg_start_compress(...);
119 while (scan lines remain to be written)
120 jpeg_write_scanlines(...);
121 jpeg_finish_compress(...);
122 Release the JPEG compression object
124 A JPEG compression object holds parameters and working state for the JPEG
125 library. We make creation/destruction of the object separate from starting
126 or finishing compression of an image; the same object can be re-used for a
127 series of image compression operations. This makes it easy to re-use the
128 same parameter settings for a sequence of images. Re-use of a JPEG object
129 also has important implications for processing abbreviated JPEG datastreams,
132 The image data to be compressed is supplied to jpeg_write_scanlines() from
133 in-memory buffers. If the application is doing file-to-file compression,
134 reading image data from the source file is the application's responsibility.
135 The library emits compressed data by calling a "data destination manager",
136 which typically will write the data into a file; but the application can
137 provide its own destination manager to do something else.
139 Similarly, the rough outline of a JPEG decompression operation is:
141 Allocate and initialize a JPEG decompression object
142 Specify the source of the compressed data (eg, a file)
143 Call jpeg_read_header() to obtain image info
144 Set parameters for decompression
145 jpeg_start_decompress(...);
146 while (scan lines remain to be read)
147 jpeg_read_scanlines(...);
148 jpeg_finish_decompress(...);
149 Release the JPEG decompression object
151 This is comparable to the compression outline except that reading the
152 datastream header is a separate step. This is helpful because information
153 about the image's size, colorspace, etc is available when the application
154 selects decompression parameters. For example, the application can choose an
155 output scaling ratio that will fit the image into the available screen size.
157 The decompression library obtains compressed data by calling a data source
158 manager, which typically will read the data from a file; but other behaviors
159 can be obtained with a custom source manager. Decompressed data is delivered
160 into in-memory buffers passed to jpeg_read_scanlines().
162 It is possible to abort an incomplete compression or decompression operation
163 by calling jpeg_abort(); or, if you do not need to retain the JPEG object,
164 simply release it by calling jpeg_destroy().
166 JPEG compression and decompression objects are two separate struct types.
167 However, they share some common fields, and certain routines such as
168 jpeg_destroy() can work on either type of object.
170 The JPEG library has no static variables: all state is in the compression
171 or decompression object. Therefore it is possible to process multiple
172 compression and decompression operations concurrently, using multiple JPEG
175 Both compression and decompression can be done in an incremental memory-to-
176 memory fashion, if suitable source/destination managers are used. See the
177 section on "I/O suspension" for more details.
186 Before diving into procedural details, it is helpful to understand the
187 image data format that the JPEG library expects or returns.
189 The standard input image format is a rectangular array of pixels, with each
190 pixel having the same number of "component" or "sample" values (color
191 channels). You must specify how many components there are and the colorspace
192 interpretation of the components. Most applications will use RGB data
193 (three components per pixel) or grayscale data (one component per pixel).
194 PLEASE NOTE THAT RGB DATA IS THREE SAMPLES PER PIXEL, GRAYSCALE ONLY ONE.
195 A remarkable number of people manage to miss this, only to find that their
196 programs don't work with grayscale JPEG files.
198 There is no provision for colormapped input. JPEG files are always full-color
199 or full grayscale (or sometimes another colorspace such as CMYK). You can
200 feed in a colormapped image by expanding it to full-color format. However
201 JPEG often doesn't work very well with source data that has been colormapped,
202 because of dithering noise. This is discussed in more detail in the JPEG FAQ
203 and the other references mentioned in the README file.
205 Pixels are stored by scanlines, with each scanline running from left to
206 right. The component values for each pixel are adjacent in the row; for
207 example, R,G,B,R,G,B,R,G,B,... for 24-bit RGB color. Each scanline is an
208 array of data type JSAMPLE --- which is typically "unsigned char", unless
209 you've changed jmorecfg.h. (You can also change the RGB pixel layout, say
210 to B,G,R order, by modifying jmorecfg.h. But see the restrictions listed in
211 that file before doing so.)
213 A 2-D array of pixels is formed by making a list of pointers to the starts of
214 scanlines; so the scanlines need not be physically adjacent in memory. Even
215 if you process just one scanline at a time, you must make a one-element
216 pointer array to conform to this structure. Pointers to JSAMPLE rows are of
217 type JSAMPROW, and the pointer to the pointer array is of type JSAMPARRAY.
219 The library accepts or supplies one or more complete scanlines per call.
220 It is not possible to process part of a row at a time. Scanlines are always
221 processed top-to-bottom. You can process an entire image in one call if you
222 have it all in memory, but usually it's simplest to process one scanline at
225 For best results, source data values should have the precision specified by
226 BITS_IN_JSAMPLE (normally 8 bits). For instance, if you choose to compress
227 data that's only 6 bits/channel, you should left-justify each value in a
228 byte before passing it to the compressor. If you need to compress data
229 that has more than 8 bits/channel, compile with BITS_IN_JSAMPLE = 12.
230 (See "Library compile-time options", later.)
233 The data format returned by the decompressor is the same in all details,
234 except that colormapped output is supported. (Again, a JPEG file is never
235 colormapped. But you can ask the decompressor to perform on-the-fly color
236 quantization to deliver colormapped output.) If you request colormapped
237 output then the returned data array contains a single JSAMPLE per pixel;
238 its value is an index into a color map. The color map is represented as
239 a 2-D JSAMPARRAY in which each row holds the values of one color component,
240 that is, colormap[i][j] is the value of the i'th color component for pixel
241 value (map index) j. Note that since the colormap indexes are stored in
242 JSAMPLEs, the maximum number of colors is limited by the size of JSAMPLE
243 (ie, at most 256 colors for an 8-bit JPEG library).
249 Here we revisit the JPEG compression outline given in the overview.
251 1. Allocate and initialize a JPEG compression object.
253 A JPEG compression object is a "struct jpeg_compress_struct". (It also has
254 a bunch of subsidiary structures which are allocated via malloc(), but the
255 application doesn't control those directly.) This struct can be just a local
256 variable in the calling routine, if a single routine is going to execute the
257 whole JPEG compression sequence. Otherwise it can be static or allocated
260 You will also need a structure representing a JPEG error handler. The part
261 of this that the library cares about is a "struct jpeg_error_mgr". If you
262 are providing your own error handler, you'll typically want to embed the
263 jpeg_error_mgr struct in a larger structure; this is discussed later under
264 "Error handling". For now we'll assume you are just using the default error
265 handler. The default error handler will print JPEG error/warning messages
266 on stderr, and it will call exit() if a fatal error occurs.
268 You must initialize the error handler structure, store a pointer to it into
269 the JPEG object's "err" field, and then call jpeg_create_compress() to
270 initialize the rest of the JPEG object.
272 Typical code for this step, if you are using the default error handler, is
274 struct jpeg_compress_struct cinfo;
275 struct jpeg_error_mgr jerr;
277 cinfo.err = jpeg_std_error(&jerr);
278 jpeg_create_compress(&cinfo);
280 jpeg_create_compress allocates a small amount of memory, so it could fail
281 if you are out of memory. In that case it will exit via the error handler;
282 that's why the error handler must be initialized first.
285 2. Specify the destination for the compressed data (eg, a file).
287 As previously mentioned, the JPEG library delivers compressed data to a
288 "data destination" module. The library includes one data destination
289 module which knows how to write to a stdio stream. You can use your own
290 destination module if you want to do something else, as discussed later.
292 If you use the standard destination module, you must open the target stdio
293 stream beforehand. Typical code for this step looks like:
297 if ((outfile = fopen(filename, "wb")) == NULL) {
298 fprintf(stderr, "can't open %s\n", filename);
301 jpeg_stdio_dest(&cinfo, outfile);
303 where the last line invokes the standard destination module.
305 WARNING: it is critical that the binary compressed data be delivered to the
306 output file unchanged. On non-Unix systems the stdio library may perform
307 newline translation or otherwise corrupt binary data. To suppress this
308 behavior, you may need to use a "b" option to fopen (as shown above), or use
309 setmode() or another routine to put the stdio stream in binary mode. See
310 cjpeg.c and djpeg.c for code that has been found to work on many systems.
312 You can select the data destination after setting other parameters (step 3),
313 if that's more convenient. You may not change the destination between
314 calling jpeg_start_compress() and jpeg_finish_compress().
317 3. Set parameters for compression, including image size & colorspace.
319 You must supply information about the source image by setting the following
320 fields in the JPEG object (cinfo structure):
322 image_width Width of image, in pixels
323 image_height Height of image, in pixels
324 input_components Number of color channels (samples per pixel)
325 in_color_space Color space of source image
327 The image dimensions are, hopefully, obvious. JPEG supports image dimensions
328 of 1 to 64K pixels in either direction. The input color space is typically
329 RGB or grayscale, and input_components is 3 or 1 accordingly. (See "Special
330 color spaces", later, for more info.) The in_color_space field must be
331 assigned one of the J_COLOR_SPACE enum constants, typically JCS_RGB or
334 JPEG has a large number of compression parameters that determine how the
335 image is encoded. Most applications don't need or want to know about all
336 these parameters. You can set all the parameters to reasonable defaults by
337 calling jpeg_set_defaults(); then, if there are particular values you want
338 to change, you can do so after that. The "Compression parameter selection"
339 section tells about all the parameters.
341 You must set in_color_space correctly before calling jpeg_set_defaults(),
342 because the defaults depend on the source image colorspace. However the
343 other three source image parameters need not be valid until you call
344 jpeg_start_compress(). There's no harm in calling jpeg_set_defaults() more
345 than once, if that happens to be convenient.
347 Typical code for a 24-bit RGB source image is
349 cinfo.image_width = Width; /* image width and height, in pixels */
350 cinfo.image_height = Height;
351 cinfo.input_components = 3; /* # of color components per pixel */
352 cinfo.in_color_space = JCS_RGB; /* colorspace of input image */
354 jpeg_set_defaults(&cinfo);
355 /* Make optional parameter settings here */
358 4. jpeg_start_compress(...);
360 After you have established the data destination and set all the necessary
361 source image info and other parameters, call jpeg_start_compress() to begin
362 a compression cycle. This will initialize internal state, allocate working
363 storage, and emit the first few bytes of the JPEG datastream header.
367 jpeg_start_compress(&cinfo, TRUE);
369 The "TRUE" parameter ensures that a complete JPEG interchange datastream
370 will be written. This is appropriate in most cases. If you think you might
371 want to use an abbreviated datastream, read the section on abbreviated
374 Once you have called jpeg_start_compress(), you may not alter any JPEG
375 parameters or other fields of the JPEG object until you have completed
376 the compression cycle.
379 5. while (scan lines remain to be written)
380 jpeg_write_scanlines(...);
382 Now write all the required image data by calling jpeg_write_scanlines()
383 one or more times. You can pass one or more scanlines in each call, up
384 to the total image height. In most applications it is convenient to pass
385 just one or a few scanlines at a time. The expected format for the passed
386 data is discussed under "Data formats", above.
388 Image data should be written in top-to-bottom scanline order. The JPEG spec
389 contains some weasel wording about how top and bottom are application-defined
390 terms (a curious interpretation of the English language...) but if you want
391 your files to be compatible with everyone else's, you WILL use top-to-bottom
392 order. If the source data must be read in bottom-to-top order, you can use
393 the JPEG library's virtual array mechanism to invert the data efficiently.
394 Examples of this can be found in the sample application cjpeg.
396 The library maintains a count of the number of scanlines written so far
397 in the next_scanline field of the JPEG object. Usually you can just use
398 this variable as the loop counter, so that the loop test looks like
399 "while (cinfo.next_scanline < cinfo.image_height)".
401 Code for this step depends heavily on the way that you store the source data.
402 example.c shows the following code for the case of a full-size 2-D source
403 array containing 3-byte RGB pixels:
405 JSAMPROW row_pointer[1]; /* pointer to a single row */
406 int row_stride; /* physical row width in buffer */
408 row_stride = image_width * 3; /* JSAMPLEs per row in image_buffer */
410 while (cinfo.next_scanline < cinfo.image_height) {
411 row_pointer[0] = & image_buffer[cinfo.next_scanline * row_stride];
412 jpeg_write_scanlines(&cinfo, row_pointer, 1);
415 jpeg_write_scanlines() returns the number of scanlines actually written.
416 This will normally be equal to the number passed in, so you can usually
417 ignore the return value. It is different in just two cases:
418 * If you try to write more scanlines than the declared image height,
419 the additional scanlines are ignored.
420 * If you use a suspending data destination manager, output buffer overrun
421 will cause the compressor to return before accepting all the passed lines.
422 This feature is discussed under "I/O suspension", below. The normal
423 stdio destination manager will NOT cause this to happen.
424 In any case, the return value is the same as the change in the value of
428 6. jpeg_finish_compress(...);
430 After all the image data has been written, call jpeg_finish_compress() to
431 complete the compression cycle. This step is ESSENTIAL to ensure that the
432 last bufferload of data is written to the data destination.
433 jpeg_finish_compress() also releases working memory associated with the JPEG
438 jpeg_finish_compress(&cinfo);
440 If using the stdio destination manager, don't forget to close the output
441 stdio stream (if necessary) afterwards.
443 If you have requested a multi-pass operating mode, such as Huffman code
444 optimization, jpeg_finish_compress() will perform the additional passes using
445 data buffered by the first pass. In this case jpeg_finish_compress() may take
446 quite a while to complete. With the default compression parameters, this will
449 It is an error to call jpeg_finish_compress() before writing the necessary
450 total number of scanlines. If you wish to abort compression, call
451 jpeg_abort() as discussed below.
453 After completing a compression cycle, you may dispose of the JPEG object
454 as discussed next, or you may use it to compress another image. In that case
455 return to step 2, 3, or 4 as appropriate. If you do not change the
456 destination manager, the new datastream will be written to the same target.
457 If you do not change any JPEG parameters, the new datastream will be written
458 with the same parameters as before. Note that you can change the input image
459 dimensions freely between cycles, but if you change the input colorspace, you
460 should call jpeg_set_defaults() to adjust for the new colorspace; and then
461 you'll need to repeat all of step 3.
464 7. Release the JPEG compression object.
466 When you are done with a JPEG compression object, destroy it by calling
467 jpeg_destroy_compress(). This will free all subsidiary memory (regardless of
468 the previous state of the object). Or you can call jpeg_destroy(), which
469 works for either compression or decompression objects --- this may be more
470 convenient if you are sharing code between compression and decompression
471 cases. (Actually, these routines are equivalent except for the declared type
472 of the passed pointer. To avoid gripes from ANSI C compilers, jpeg_destroy()
473 should be passed a j_common_ptr.)
475 If you allocated the jpeg_compress_struct structure from malloc(), freeing
476 it is your responsibility --- jpeg_destroy() won't. Ditto for the error
481 jpeg_destroy_compress(&cinfo);
486 If you decide to abort a compression cycle before finishing, you can clean up
487 in either of two ways:
489 * If you don't need the JPEG object any more, just call
490 jpeg_destroy_compress() or jpeg_destroy() to release memory. This is
491 legitimate at any point after calling jpeg_create_compress() --- in fact,
492 it's safe even if jpeg_create_compress() fails.
494 * If you want to re-use the JPEG object, call jpeg_abort_compress(), or call
495 jpeg_abort() which works on both compression and decompression objects.
496 This will return the object to an idle state, releasing any working memory.
497 jpeg_abort() is allowed at any time after successful object creation.
499 Note that cleaning up the data destination, if required, is your
500 responsibility; neither of these routines will call term_destination().
501 (See "Compressed data handling", below, for more about that.)
503 jpeg_destroy() and jpeg_abort() are the only safe calls to make on a JPEG
504 object that has reported an error by calling error_exit (see "Error handling"
505 for more info). The internal state of such an object is likely to be out of
506 whack. Either of these two routines will return the object to a known state.
509 Decompression details
510 ---------------------
512 Here we revisit the JPEG decompression outline given in the overview.
514 1. Allocate and initialize a JPEG decompression object.
516 This is just like initialization for compression, as discussed above,
517 except that the object is a "struct jpeg_decompress_struct" and you
518 call jpeg_create_decompress(). Error handling is exactly the same.
522 struct jpeg_decompress_struct cinfo;
523 struct jpeg_error_mgr jerr;
525 cinfo.err = jpeg_std_error(&jerr);
526 jpeg_create_decompress(&cinfo);
528 (Both here and in the IJG code, we usually use variable name "cinfo" for
529 both compression and decompression objects.)
532 2. Specify the source of the compressed data (eg, a file).
534 As previously mentioned, the JPEG library reads compressed data from a "data
535 source" module. The library includes one data source module which knows how
536 to read from a stdio stream. You can use your own source module if you want
537 to do something else, as discussed later.
539 If you use the standard source module, you must open the source stdio stream
540 beforehand. Typical code for this step looks like:
544 if ((infile = fopen(filename, "rb")) == NULL) {
545 fprintf(stderr, "can't open %s\n", filename);
548 jpeg_stdio_src(&cinfo, infile);
550 where the last line invokes the standard source module.
552 WARNING: it is critical that the binary compressed data be read unchanged.
553 On non-Unix systems the stdio library may perform newline translation or
554 otherwise corrupt binary data. To suppress this behavior, you may need to use
555 a "b" option to fopen (as shown above), or use setmode() or another routine to
556 put the stdio stream in binary mode. See cjpeg.c and djpeg.c for code that
557 has been found to work on many systems.
559 You may not change the data source between calling jpeg_read_header() and
560 jpeg_finish_decompress(). If you wish to read a series of JPEG images from
561 a single source file, you should repeat the jpeg_read_header() to
562 jpeg_finish_decompress() sequence without reinitializing either the JPEG
563 object or the data source module; this prevents buffered input data from
567 3. Call jpeg_read_header() to obtain image info.
569 Typical code for this step is just
571 jpeg_read_header(&cinfo, TRUE);
573 This will read the source datastream header markers, up to the beginning
574 of the compressed data proper. On return, the image dimensions and other
575 info have been stored in the JPEG object. The application may wish to
576 consult this information before selecting decompression parameters.
578 More complex code is necessary if
579 * A suspending data source is used --- in that case jpeg_read_header()
580 may return before it has read all the header data. See "I/O suspension",
581 below. The normal stdio source manager will NOT cause this to happen.
582 * Abbreviated JPEG files are to be processed --- see the section on
583 abbreviated datastreams. Standard applications that deal only in
584 interchange JPEG files need not be concerned with this case either.
586 It is permissible to stop at this point if you just wanted to find out the
587 image dimensions and other header info for a JPEG file. In that case,
588 call jpeg_destroy() when you are done with the JPEG object, or call
589 jpeg_abort() to return it to an idle state before selecting a new data
590 source and reading another header.
593 4. Set parameters for decompression.
595 jpeg_read_header() sets appropriate default decompression parameters based on
596 the properties of the image (in particular, its colorspace). However, you
597 may well want to alter these defaults before beginning the decompression.
598 For example, the default is to produce full color output from a color file.
599 If you want colormapped output you must ask for it. Other options allow the
600 returned image to be scaled and allow various speed/quality tradeoffs to be
601 selected. "Decompression parameter selection", below, gives details.
603 If the defaults are appropriate, nothing need be done at this step.
605 Note that all default values are set by each call to jpeg_read_header().
606 If you reuse a decompression object, you cannot expect your parameter
607 settings to be preserved across cycles, as you can for compression.
608 You must set desired parameter values each time.
611 5. jpeg_start_decompress(...);
613 Once the parameter values are satisfactory, call jpeg_start_decompress() to
614 begin decompression. This will initialize internal state, allocate working
615 memory, and prepare for returning data.
619 jpeg_start_decompress(&cinfo);
621 If you have requested a multi-pass operating mode, such as 2-pass color
622 quantization, jpeg_start_decompress() will do everything needed before data
623 output can begin. In this case jpeg_start_decompress() may take quite a while
624 to complete. With a single-scan (non progressive) JPEG file and default
625 decompression parameters, this will not happen; jpeg_start_decompress() will
628 After this call, the final output image dimensions, including any requested
629 scaling, are available in the JPEG object; so is the selected colormap, if
630 colormapped output has been requested. Useful fields include
632 output_width image width and height, as scaled
634 out_color_components # of color components in out_color_space
635 output_components # of color components returned per pixel
636 colormap the selected colormap, if any
637 actual_number_of_colors number of entries in colormap
639 output_components is 1 (a colormap index) when quantizing colors; otherwise it
640 equals out_color_components. It is the number of JSAMPLE values that will be
641 emitted per pixel in the output arrays.
643 Typically you will need to allocate data buffers to hold the incoming image.
644 You will need output_width * output_components JSAMPLEs per scanline in your
645 output buffer, and a total of output_height scanlines will be returned.
647 Note: if you are using the JPEG library's internal memory manager to allocate
648 data buffers (as djpeg does), then the manager's protocol requires that you
649 request large buffers *before* calling jpeg_start_decompress(). This is a
650 little tricky since the output_XXX fields are not normally valid then. You
651 can make them valid by calling jpeg_calc_output_dimensions() after setting the
652 relevant parameters (scaling, output color space, and quantization flag).
655 6. while (scan lines remain to be read)
656 jpeg_read_scanlines(...);
658 Now you can read the decompressed image data by calling jpeg_read_scanlines()
659 one or more times. At each call, you pass in the maximum number of scanlines
660 to be read (ie, the height of your working buffer); jpeg_read_scanlines()
661 will return up to that many lines. The return value is the number of lines
662 actually read. The format of the returned data is discussed under "Data
663 formats", above. Don't forget that grayscale and color JPEGs will return
664 different data formats!
666 Image data is returned in top-to-bottom scanline order. If you must write
667 out the image in bottom-to-top order, you can use the JPEG library's virtual
668 array mechanism to invert the data efficiently. Examples of this can be
669 found in the sample application djpeg.
671 The library maintains a count of the number of scanlines returned so far
672 in the output_scanline field of the JPEG object. Usually you can just use
673 this variable as the loop counter, so that the loop test looks like
674 "while (cinfo.output_scanline < cinfo.output_height)". (Note that the test
675 should NOT be against image_height, unless you never use scaling. The
676 image_height field is the height of the original unscaled image.)
677 The return value always equals the change in the value of output_scanline.
679 If you don't use a suspending data source, it is safe to assume that
680 jpeg_read_scanlines() reads at least one scanline per call, until the
681 bottom of the image has been reached.
683 If you use a buffer larger than one scanline, it is NOT safe to assume that
684 jpeg_read_scanlines() fills it. (The current implementation returns only a
685 few scanlines per call, no matter how large a buffer you pass.) So you must
686 always provide a loop that calls jpeg_read_scanlines() repeatedly until the
687 whole image has been read.
690 7. jpeg_finish_decompress(...);
692 After all the image data has been read, call jpeg_finish_decompress() to
693 complete the decompression cycle. This causes working memory associated
694 with the JPEG object to be released.
698 jpeg_finish_decompress(&cinfo);
700 If using the stdio source manager, don't forget to close the source stdio
703 It is an error to call jpeg_finish_decompress() before reading the correct
704 total number of scanlines. If you wish to abort decompression, call
705 jpeg_abort() as discussed below.
707 After completing a decompression cycle, you may dispose of the JPEG object as
708 discussed next, or you may use it to decompress another image. In that case
709 return to step 2 or 3 as appropriate. If you do not change the source
710 manager, the next image will be read from the same source.
713 8. Release the JPEG decompression object.
715 When you are done with a JPEG decompression object, destroy it by calling
716 jpeg_destroy_decompress() or jpeg_destroy(). The previous discussion of
717 destroying compression objects applies here too.
721 jpeg_destroy_decompress(&cinfo);
726 You can abort a decompression cycle by calling jpeg_destroy_decompress() or
727 jpeg_destroy() if you don't need the JPEG object any more, or
728 jpeg_abort_decompress() or jpeg_abort() if you want to reuse the object.
729 The previous discussion of aborting compression cycles applies here too.
732 Mechanics of usage: include files, linking, etc
733 -----------------------------------------------
735 Applications using the JPEG library should include the header file jpeglib.h
736 to obtain declarations of data types and routines. Before including
737 jpeglib.h, include system headers that define at least the typedefs FILE and
738 size_t. On ANSI-conforming systems, including <stdio.h> is sufficient; on
739 older Unix systems, you may need <sys/types.h> to define size_t.
741 If the application needs to refer to individual JPEG library error codes, also
742 include jerror.h to define those symbols.
744 jpeglib.h indirectly includes the files jconfig.h and jmorecfg.h. If you are
745 installing the JPEG header files in a system directory, you will want to
746 install all four files: jpeglib.h, jerror.h, jconfig.h, jmorecfg.h.
748 The most convenient way to include the JPEG code into your executable program
749 is to prepare a library file ("libjpeg.a", or a corresponding name on non-Unix
750 machines) and reference it at your link step. If you use only half of the
751 library (only compression or only decompression), only that much code will be
752 included from the library, unless your linker is hopelessly brain-damaged.
753 The supplied makefiles build libjpeg.a automatically (see install.txt).
755 While you can build the JPEG library as a shared library if the whim strikes
756 you, we don't really recommend it. The trouble with shared libraries is that
757 at some point you'll probably try to substitute a new version of the library
758 without recompiling the calling applications. That generally doesn't work
759 because the parameter struct declarations usually change with each new
760 version. In other words, the library's API is *not* guaranteed binary
761 compatible across versions; we only try to ensure source-code compatibility.
762 (In hindsight, it might have been smarter to hide the parameter structs from
763 applications and introduce a ton of access functions instead. Too late now,
766 It may be worth pointing out that the core JPEG library does not actually
767 require the stdio library: only the default source/destination managers and
768 error handler need it. You can use the library in a stdio-less environment
769 if you replace those modules and use jmemnobs.c (or another memory manager of
770 your own devising). More info about the minimum system library requirements
771 may be found in jinclude.h.
777 Compression parameter selection
778 -------------------------------
780 This section describes all the optional parameters you can set for JPEG
781 compression, as well as the "helper" routines provided to assist in this
782 task. Proper setting of some parameters requires detailed understanding
783 of the JPEG standard; if you don't know what a parameter is for, it's best
784 not to mess with it! See REFERENCES in the README file for pointers to
785 more info about JPEG.
787 It's a good idea to call jpeg_set_defaults() first, even if you plan to set
788 all the parameters; that way your code is more likely to work with future JPEG
789 libraries that have additional parameters. For the same reason, we recommend
790 you use a helper routine where one is provided, in preference to twiddling
791 cinfo fields directly.
793 The helper routines are:
795 jpeg_set_defaults (j_compress_ptr cinfo)
796 This routine sets all JPEG parameters to reasonable defaults, using
797 only the input image's color space (field in_color_space, which must
798 already be set in cinfo). Many applications will only need to use
799 this routine and perhaps jpeg_set_quality().
801 jpeg_set_colorspace (j_compress_ptr cinfo, J_COLOR_SPACE colorspace)
802 Sets the JPEG file's colorspace (field jpeg_color_space) as specified,
803 and sets other color-space-dependent parameters appropriately. See
804 "Special color spaces", below, before using this. A large number of
805 parameters, including all per-component parameters, are set by this
806 routine; if you want to twiddle individual parameters you should call
807 jpeg_set_colorspace() before rather than after.
809 jpeg_default_colorspace (j_compress_ptr cinfo)
810 Selects an appropriate JPEG colorspace based on cinfo->in_color_space,
811 and calls jpeg_set_colorspace(). This is actually a subroutine of
812 jpeg_set_defaults(). It's broken out in case you want to change
813 just the colorspace-dependent JPEG parameters.
815 jpeg_set_quality (j_compress_ptr cinfo, int quality, boolean force_baseline)
816 Constructs JPEG quantization tables appropriate for the indicated
817 quality setting. The quality value is expressed on the 0..100 scale
818 recommended by IJG (cjpeg's "-quality" switch uses this routine).
819 Note that the exact mapping from quality values to tables may change
820 in future IJG releases as more is learned about DCT quantization.
821 If the force_baseline parameter is TRUE, then the quantization table
822 entries are constrained to the range 1..255 for full JPEG baseline
823 compatibility. In the current implementation, this only makes a
824 difference for quality settings below 25, and it effectively prevents
825 very small/low quality files from being generated. The IJG decoder
826 is capable of reading the non-baseline files generated at low quality
827 settings when force_baseline is FALSE, but other decoders may not be.
829 jpeg_set_linear_quality (j_compress_ptr cinfo, int scale_factor,
830 boolean force_baseline)
831 Same as jpeg_set_quality() except that the generated tables are the
832 sample tables given in the JPEC spec section K.1, multiplied by the
833 specified scale factor (which is expressed as a percentage; thus
834 scale_factor = 100 reproduces the spec's tables). Note that larger
835 scale factors give lower quality. This entry point is useful for
836 conforming to the Adobe PostScript DCT conventions, but we do not
837 recommend linear scaling as a user-visible quality scale otherwise.
838 force_baseline again constrains the computed table entries to 1..255.
840 int jpeg_quality_scaling (int quality)
841 Converts a value on the IJG-recommended quality scale to a linear
842 scaling percentage. Note that this routine may change or go away
843 in future releases --- IJG may choose to adopt a scaling method that
844 can't be expressed as a simple scalar multiplier, in which case the
845 premise of this routine collapses. Caveat user.
847 jpeg_default_qtables (j_compress_ptr cinfo, boolean force_baseline)
848 [libjpeg v7+ API/ABI emulation only]
849 Set default quantization tables with linear q_scale_factor[] values
852 jpeg_add_quant_table (j_compress_ptr cinfo, int which_tbl,
853 const unsigned int *basic_table,
854 int scale_factor, boolean force_baseline)
855 Allows an arbitrary quantization table to be created. which_tbl
856 indicates which table slot to fill. basic_table points to an array
857 of 64 unsigned ints given in normal array order. These values are
858 multiplied by scale_factor/100 and then clamped to the range 1..65535
859 (or to 1..255 if force_baseline is TRUE).
860 CAUTION: prior to library version 6a, jpeg_add_quant_table expected
861 the basic table to be given in JPEG zigzag order. If you need to
862 write code that works with either older or newer versions of this
863 routine, you must check the library version number. Something like
864 "#if JPEG_LIB_VERSION >= 61" is the right test.
866 jpeg_simple_progression (j_compress_ptr cinfo)
867 Generates a default scan script for writing a progressive-JPEG file.
868 This is the recommended method of creating a progressive file,
869 unless you want to make a custom scan sequence. You must ensure that
870 the JPEG color space is set correctly before calling this routine.
873 Compression parameters (cinfo fields) include:
875 J_DCT_METHOD dct_method
876 Selects the algorithm used for the DCT step. Choices are:
877 JDCT_ISLOW: slow but accurate integer algorithm
878 JDCT_IFAST: faster, less accurate integer method
879 JDCT_FLOAT: floating-point method
880 JDCT_DEFAULT: default method (normally JDCT_ISLOW)
881 JDCT_FASTEST: fastest method (normally JDCT_IFAST)
882 In libjpeg-turbo, JDCT_IFAST is generally about 5-15% faster than
883 JDCT_ISLOW when using the x86/x86-64 SIMD extensions (results may vary
884 with other SIMD implementations, or when using libjpeg-turbo without
885 SIMD extensions.) For quality levels of 90 and below, there should be
886 little or no perceptible difference between the two algorithms. For
887 quality levels above 90, however, the difference between JDCT_IFAST and
888 JDCT_ISLOW becomes more pronounced. With quality=97, for instance,
889 JDCT_IFAST incurs generally about a 1-3 dB loss (in PSNR) relative to
890 JDCT_ISLOW, but this can be larger for some images. Do not use
891 JDCT_IFAST with quality levels above 97. The algorithm often
892 degenerates at quality=98 and above and can actually produce a more
893 lossy image than if lower quality levels had been used. Also, in
894 libjpeg-turbo, JDCT_IFAST is not fully accelerated for quality levels
895 above 97, so it will be slower than JDCT_ISLOW. JDCT_FLOAT is mainly a
896 legacy feature. It does not produce significantly more accurate
897 results than the ISLOW method, and it is much slower. The FLOAT method
898 may also give different results on different machines due to varying
899 roundoff behavior, whereas the integer methods should give the same
900 results on all machines.
902 J_COLOR_SPACE jpeg_color_space
904 The JPEG color space and corresponding number of components; see
905 "Special color spaces", below, for more info. We recommend using
906 jpeg_set_color_space() if you want to change these.
908 boolean optimize_coding
909 TRUE causes the compressor to compute optimal Huffman coding tables
910 for the image. This requires an extra pass over the data and
911 therefore costs a good deal of space and time. The default is
912 FALSE, which tells the compressor to use the supplied or default
913 Huffman tables. In most cases optimal tables save only a few percent
914 of file size compared to the default tables. Note that when this is
915 TRUE, you need not supply Huffman tables at all, and any you do
916 supply will be overwritten.
918 unsigned int restart_interval
920 To emit restart markers in the JPEG file, set one of these nonzero.
921 Set restart_interval to specify the exact interval in MCU blocks.
922 Set restart_in_rows to specify the interval in MCU rows. (If
923 restart_in_rows is not 0, then restart_interval is set after the
924 image width in MCUs is computed.) Defaults are zero (no restarts).
925 One restart marker per MCU row is often a good choice.
926 NOTE: the overhead of restart markers is higher in grayscale JPEG
927 files than in color files, and MUCH higher in progressive JPEGs.
928 If you use restarts, you may want to use larger intervals in those
931 const jpeg_scan_info * scan_info
933 By default, scan_info is NULL; this causes the compressor to write a
934 single-scan sequential JPEG file. If not NULL, scan_info points to
935 an array of scan definition records of length num_scans. The
936 compressor will then write a JPEG file having one scan for each scan
937 definition record. This is used to generate noninterleaved or
938 progressive JPEG files. The library checks that the scan array
939 defines a valid JPEG scan sequence. (jpeg_simple_progression creates
940 a suitable scan definition array for progressive JPEG.) This is
941 discussed further under "Progressive JPEG support".
944 If non-zero, the input image is smoothed; the value should be 1 for
945 minimal smoothing to 100 for maximum smoothing. Consult jcsample.c
946 for details of the smoothing algorithm. The default is zero.
948 boolean write_JFIF_header
949 If TRUE, a JFIF APP0 marker is emitted. jpeg_set_defaults() and
950 jpeg_set_colorspace() set this TRUE if a JFIF-legal JPEG color space
951 (ie, YCbCr or grayscale) is selected, otherwise FALSE.
953 UINT8 JFIF_major_version
954 UINT8 JFIF_minor_version
955 The version number to be written into the JFIF marker.
956 jpeg_set_defaults() initializes the version to 1.01 (major=minor=1).
957 You should set it to 1.02 (major=1, minor=2) if you plan to write
958 any JFIF 1.02 extension markers.
963 The resolution information to be written into the JFIF marker;
964 not used otherwise. density_unit may be 0 for unknown,
965 1 for dots/inch, or 2 for dots/cm. The default values are 0,1,1
966 indicating square pixels of unknown size.
968 boolean write_Adobe_marker
969 If TRUE, an Adobe APP14 marker is emitted. jpeg_set_defaults() and
970 jpeg_set_colorspace() set this TRUE if JPEG color space RGB, CMYK,
971 or YCCK is selected, otherwise FALSE. It is generally a bad idea
972 to set both write_JFIF_header and write_Adobe_marker. In fact,
973 you probably shouldn't change the default settings at all --- the
974 default behavior ensures that the JPEG file's color space can be
975 recognized by the decoder.
977 JQUANT_TBL * quant_tbl_ptrs[NUM_QUANT_TBLS]
978 Pointers to coefficient quantization tables, one per table slot,
979 or NULL if no table is defined for a slot. Usually these should
980 be set via one of the above helper routines; jpeg_add_quant_table()
981 is general enough to define any quantization table. The other
982 routines will set up table slot 0 for luminance quality and table
983 slot 1 for chrominance.
985 int q_scale_factor[NUM_QUANT_TBLS]
986 [libjpeg v7+ API/ABI emulation only]
987 Linear quantization scaling factors (0-100, default 100)
988 for use with jpeg_default_qtables().
989 See rdswitch.c and cjpeg.c for an example of usage.
990 Note that the q_scale_factor[] values use "linear" scales, so JPEG
991 quality levels chosen by the user must be converted to these scales
992 using jpeg_quality_scaling(). Here is an example that corresponds to
993 cjpeg -quality 90,70:
995 jpeg_set_defaults(cinfo);
997 /* Set luminance quality 90. */
998 cinfo->q_scale_factor[0] = jpeg_quality_scaling(90);
999 /* Set chrominance quality 70. */
1000 cinfo->q_scale_factor[1] = jpeg_quality_scaling(70);
1002 jpeg_default_qtables(cinfo, force_baseline);
1004 CAUTION: Setting separate quality levels for chrominance and luminance
1005 is mainly only useful if chrominance subsampling is disabled. 2x2
1006 chrominance subsampling (AKA "4:2:0") is the default, but you can
1007 explicitly disable subsampling as follows:
1009 cinfo->comp_info[0].v_samp_factor = 1;
1010 cinfo->comp_info[0].h_samp_factor = 1;
1012 JHUFF_TBL * dc_huff_tbl_ptrs[NUM_HUFF_TBLS]
1013 JHUFF_TBL * ac_huff_tbl_ptrs[NUM_HUFF_TBLS]
1014 Pointers to Huffman coding tables, one per table slot, or NULL if
1015 no table is defined for a slot. Slots 0 and 1 are filled with the
1016 JPEG sample tables by jpeg_set_defaults(). If you need to allocate
1017 more table structures, jpeg_alloc_huff_table() may be used.
1018 Note that optimal Huffman tables can be computed for an image
1019 by setting optimize_coding, as discussed above; there's seldom
1020 any need to mess with providing your own Huffman tables.
1023 [libjpeg v7+ API/ABI emulation only]
1024 The actual dimensions of the JPEG image that will be written to the file are
1025 given by the following fields. These are computed from the input image
1026 dimensions and the compression parameters by jpeg_start_compress(). You can
1027 also call jpeg_calc_jpeg_dimensions() to obtain the values that will result
1028 from the current parameter settings. This can be useful if you are trying
1029 to pick a scaling ratio that will get close to a desired target size.
1031 JDIMENSION jpeg_width Actual dimensions of output image.
1032 JDIMENSION jpeg_height
1035 Per-component parameters are stored in the struct cinfo.comp_info[i] for
1036 component number i. Note that components here refer to components of the
1037 JPEG color space, *not* the source image color space. A suitably large
1038 comp_info[] array is allocated by jpeg_set_defaults(); if you choose not
1039 to use that routine, it's up to you to allocate the array.
1042 The one-byte identifier code to be recorded in the JPEG file for
1043 this component. For the standard color spaces, we recommend you
1044 leave the default values alone.
1048 Horizontal and vertical sampling factors for the component; must
1049 be 1..4 according to the JPEG standard. Note that larger sampling
1050 factors indicate a higher-resolution component; many people find
1051 this behavior quite unintuitive. The default values are 2,2 for
1052 luminance components and 1,1 for chrominance components, except
1053 for grayscale where 1,1 is used.
1056 Quantization table number for component. The default value is
1057 0 for luminance components and 1 for chrominance components.
1061 DC and AC entropy coding table numbers. The default values are
1062 0 for luminance components and 1 for chrominance components.
1065 Must equal the component's index in comp_info[]. (Beginning in
1066 release v6, the compressor library will fill this in automatically;
1070 Decompression parameter selection
1071 ---------------------------------
1073 Decompression parameter selection is somewhat simpler than compression
1074 parameter selection, since all of the JPEG internal parameters are
1075 recorded in the source file and need not be supplied by the application.
1076 (Unless you are working with abbreviated files, in which case see
1077 "Abbreviated datastreams", below.) Decompression parameters control
1078 the postprocessing done on the image to deliver it in a format suitable
1079 for the application's use. Many of the parameters control speed/quality
1080 tradeoffs, in which faster decompression may be obtained at the price of
1081 a poorer-quality image. The defaults select the highest quality (slowest)
1084 The following fields in the JPEG object are set by jpeg_read_header() and
1085 may be useful to the application in choosing decompression parameters:
1087 JDIMENSION image_width Width and height of image
1088 JDIMENSION image_height
1089 int num_components Number of color components
1090 J_COLOR_SPACE jpeg_color_space Colorspace of image
1091 boolean saw_JFIF_marker TRUE if a JFIF APP0 marker was seen
1092 UINT8 JFIF_major_version Version information from JFIF marker
1093 UINT8 JFIF_minor_version
1094 UINT8 density_unit Resolution data from JFIF marker
1097 boolean saw_Adobe_marker TRUE if an Adobe APP14 marker was seen
1098 UINT8 Adobe_transform Color transform code from Adobe marker
1100 The JPEG color space, unfortunately, is something of a guess since the JPEG
1101 standard proper does not provide a way to record it. In practice most files
1102 adhere to the JFIF or Adobe conventions, and the decoder will recognize these
1103 correctly. See "Special color spaces", below, for more info.
1106 The decompression parameters that determine the basic properties of the
1109 J_COLOR_SPACE out_color_space
1110 Output color space. jpeg_read_header() sets an appropriate default
1111 based on jpeg_color_space; typically it will be RGB or grayscale.
1112 The application can change this field to request output in a different
1113 colorspace. For example, set it to JCS_GRAYSCALE to get grayscale
1114 output from a color file. (This is useful for previewing: grayscale
1115 output is faster than full color since the color components need not
1116 be processed.) Note that not all possible color space transforms are
1117 currently implemented; you may need to extend jdcolor.c if you want an
1120 unsigned int scale_num, scale_denom
1121 Scale the image by the fraction scale_num/scale_denom. Default is
1122 1/1, or no scaling. Currently, the only supported scaling ratios
1123 are M/8 with all M from 1 to 16, or any reduced fraction thereof (such
1124 as 1/2, 3/4, etc.) (The library design allows for arbitrary
1125 scaling ratios but this is not likely to be implemented any time soon.)
1126 Smaller scaling ratios permit significantly faster decoding since
1127 fewer pixels need be processed and a simpler IDCT method can be used.
1129 boolean quantize_colors
1130 If set TRUE, colormapped output will be delivered. Default is FALSE,
1131 meaning that full-color output will be delivered.
1133 The next three parameters are relevant only if quantize_colors is TRUE.
1135 int desired_number_of_colors
1136 Maximum number of colors to use in generating a library-supplied color
1137 map (the actual number of colors is returned in a different field).
1138 Default 256. Ignored when the application supplies its own color map.
1140 boolean two_pass_quantize
1141 If TRUE, an extra pass over the image is made to select a custom color
1142 map for the image. This usually looks a lot better than the one-size-
1143 fits-all colormap that is used otherwise. Default is TRUE. Ignored
1144 when the application supplies its own color map.
1146 J_DITHER_MODE dither_mode
1147 Selects color dithering method. Supported values are:
1148 JDITHER_NONE no dithering: fast, very low quality
1149 JDITHER_ORDERED ordered dither: moderate speed and quality
1150 JDITHER_FS Floyd-Steinberg dither: slow, high quality
1151 Default is JDITHER_FS. (At present, ordered dither is implemented
1152 only in the single-pass, standard-colormap case. If you ask for
1153 ordered dither when two_pass_quantize is TRUE or when you supply
1154 an external color map, you'll get F-S dithering.)
1156 When quantize_colors is TRUE, the target color map is described by the next
1157 two fields. colormap is set to NULL by jpeg_read_header(). The application
1158 can supply a color map by setting colormap non-NULL and setting
1159 actual_number_of_colors to the map size. Otherwise, jpeg_start_decompress()
1160 selects a suitable color map and sets these two fields itself.
1161 [Implementation restriction: at present, an externally supplied colormap is
1162 only accepted for 3-component output color spaces.]
1165 The color map, represented as a 2-D pixel array of out_color_components
1166 rows and actual_number_of_colors columns. Ignored if not quantizing.
1167 CAUTION: if the JPEG library creates its own colormap, the storage
1168 pointed to by this field is released by jpeg_finish_decompress().
1169 Copy the colormap somewhere else first, if you want to save it.
1171 int actual_number_of_colors
1172 The number of colors in the color map.
1174 Additional decompression parameters that the application may set include:
1176 J_DCT_METHOD dct_method
1177 Selects the algorithm used for the DCT step. Choices are:
1178 JDCT_ISLOW: slow but accurate integer algorithm
1179 JDCT_IFAST: faster, less accurate integer method
1180 JDCT_FLOAT: floating-point method
1181 JDCT_DEFAULT: default method (normally JDCT_ISLOW)
1182 JDCT_FASTEST: fastest method (normally JDCT_IFAST)
1183 In libjpeg-turbo, JDCT_IFAST is generally about 5-15% faster than
1184 JDCT_ISLOW when using the x86/x86-64 SIMD extensions (results may vary
1185 with other SIMD implementations, or when using libjpeg-turbo without
1186 SIMD extensions.) If the JPEG image was compressed using a quality
1187 level of 85 or below, then there should be little or no perceptible
1188 difference between the two algorithms. When decompressing images that
1189 were compressed using quality levels above 85, however, the difference
1190 between JDCT_IFAST and JDCT_ISLOW becomes more pronounced. With images
1191 compressed using quality=97, for instance, JDCT_IFAST incurs generally
1192 about a 4-6 dB loss (in PSNR) relative to JDCT_ISLOW, but this can be
1193 larger for some images. If you can avoid it, do not use JDCT_IFAST
1194 when decompressing images that were compressed using quality levels
1195 above 97. The algorithm often degenerates for such images and can
1196 actually produce a more lossy output image than if the JPEG image had
1197 been compressed using lower quality levels. JDCT_FLOAT is mainly a
1198 legacy feature. It does not produce significantly more accurate
1199 results than the ISLOW method, and it is much slower. The FLOAT method
1200 may also give different results on different machines due to varying
1201 roundoff behavior, whereas the integer methods should give the same
1202 results on all machines.
1204 boolean do_fancy_upsampling
1205 If TRUE, do careful upsampling of chroma components. If FALSE,
1206 a faster but sloppier method is used. Default is TRUE. The visual
1207 impact of the sloppier method is often very small.
1209 boolean do_block_smoothing
1210 If TRUE, interblock smoothing is applied in early stages of decoding
1211 progressive JPEG files; if FALSE, not. Default is TRUE. Early
1212 progression stages look "fuzzy" with smoothing, "blocky" without.
1213 In any case, block smoothing ceases to be applied after the first few
1214 AC coefficients are known to full accuracy, so it is relevant only
1215 when using buffered-image mode for progressive images.
1217 boolean enable_1pass_quant
1218 boolean enable_external_quant
1219 boolean enable_2pass_quant
1220 These are significant only in buffered-image mode, which is
1221 described in its own section below.
1224 The output image dimensions are given by the following fields. These are
1225 computed from the source image dimensions and the decompression parameters
1226 by jpeg_start_decompress(). You can also call jpeg_calc_output_dimensions()
1227 to obtain the values that will result from the current parameter settings.
1228 This can be useful if you are trying to pick a scaling ratio that will get
1229 close to a desired target size. It's also important if you are using the
1230 JPEG library's memory manager to allocate output buffer space, because you
1231 are supposed to request such buffers *before* jpeg_start_decompress().
1233 JDIMENSION output_width Actual dimensions of output image.
1234 JDIMENSION output_height
1235 int out_color_components Number of color components in out_color_space.
1236 int output_components Number of color components returned.
1237 int rec_outbuf_height Recommended height of scanline buffer.
1239 When quantizing colors, output_components is 1, indicating a single color map
1240 index per pixel. Otherwise it equals out_color_components. The output arrays
1241 are required to be output_width * output_components JSAMPLEs wide.
1243 rec_outbuf_height is the recommended minimum height (in scanlines) of the
1244 buffer passed to jpeg_read_scanlines(). If the buffer is smaller, the
1245 library will still work, but time will be wasted due to unnecessary data
1246 copying. In high-quality modes, rec_outbuf_height is always 1, but some
1247 faster, lower-quality modes set it to larger values (typically 2 to 4).
1248 If you are going to ask for a high-speed processing mode, you may as well
1249 go to the trouble of honoring rec_outbuf_height so as to avoid data copying.
1250 (An output buffer larger than rec_outbuf_height lines is OK, but won't
1251 provide any material speed improvement over that height.)
1254 Special color spaces
1255 --------------------
1257 The JPEG standard itself is "color blind" and doesn't specify any particular
1258 color space. It is customary to convert color data to a luminance/chrominance
1259 color space before compressing, since this permits greater compression. The
1260 existing de-facto JPEG file format standards specify YCbCr or grayscale data
1261 (JFIF), or grayscale, RGB, YCbCr, CMYK, or YCCK (Adobe). For special
1262 applications such as multispectral images, other color spaces can be used,
1263 but it must be understood that such files will be unportable.
1265 The JPEG library can handle the most common colorspace conversions (namely
1266 RGB <=> YCbCr and CMYK <=> YCCK). It can also deal with data of an unknown
1267 color space, passing it through without conversion. If you deal extensively
1268 with an unusual color space, you can easily extend the library to understand
1269 additional color spaces and perform appropriate conversions.
1271 For compression, the source data's color space is specified by field
1272 in_color_space. This is transformed to the JPEG file's color space given
1273 by jpeg_color_space. jpeg_set_defaults() chooses a reasonable JPEG color
1274 space depending on in_color_space, but you can override this by calling
1275 jpeg_set_colorspace(). Of course you must select a supported transformation.
1276 jccolor.c currently supports the following transformations:
1281 plus the null transforms: GRAYSCALE => GRAYSCALE, RGB => RGB,
1282 YCbCr => YCbCr, CMYK => CMYK, YCCK => YCCK, and UNKNOWN => UNKNOWN.
1284 The de-facto file format standards (JFIF and Adobe) specify APPn markers that
1285 indicate the color space of the JPEG file. It is important to ensure that
1286 these are written correctly, or omitted if the JPEG file's color space is not
1287 one of the ones supported by the de-facto standards. jpeg_set_colorspace()
1288 will set the compression parameters to include or omit the APPn markers
1289 properly, so long as it is told the truth about the JPEG color space.
1290 For example, if you are writing some random 3-component color space without
1291 conversion, don't try to fake out the library by setting in_color_space and
1292 jpeg_color_space to JCS_YCbCr; use JCS_UNKNOWN. You may want to write an
1293 APPn marker of your own devising to identify the colorspace --- see "Special
1296 When told that the color space is UNKNOWN, the library will default to using
1297 luminance-quality compression parameters for all color components. You may
1298 well want to change these parameters. See the source code for
1299 jpeg_set_colorspace(), in jcparam.c, for details.
1301 For decompression, the JPEG file's color space is given in jpeg_color_space,
1302 and this is transformed to the output color space out_color_space.
1303 jpeg_read_header's setting of jpeg_color_space can be relied on if the file
1304 conforms to JFIF or Adobe conventions, but otherwise it is no better than a
1305 guess. If you know the JPEG file's color space for certain, you can override
1306 jpeg_read_header's guess by setting jpeg_color_space. jpeg_read_header also
1307 selects a default output color space based on (its guess of) jpeg_color_space;
1308 set out_color_space to override this. Again, you must select a supported
1309 transformation. jdcolor.c currently supports
1315 as well as the null transforms. (Since GRAYSCALE=>RGB is provided, an
1316 application can force grayscale JPEGs to look like color JPEGs if it only
1317 wants to handle one case.)
1319 The two-pass color quantizer, jquant2.c, is specialized to handle RGB data
1320 (it weights distances appropriately for RGB colors). You'll need to modify
1321 the code if you want to use it for non-RGB output color spaces. Note that
1322 jquant2.c is used to map to an application-supplied colormap as well as for
1323 the normal two-pass colormap selection process.
1325 CAUTION: it appears that Adobe Photoshop writes inverted data in CMYK JPEG
1326 files: 0 represents 100% ink coverage, rather than 0% ink as you'd expect.
1327 This is arguably a bug in Photoshop, but if you need to work with Photoshop
1328 CMYK files, you will have to deal with it in your application. We cannot
1329 "fix" this in the library by inverting the data during the CMYK<=>YCCK
1330 transform, because that would break other applications, notably Ghostscript.
1331 Photoshop versions prior to 3.0 write EPS files containing JPEG-encoded CMYK
1332 data in the same inverted-YCCK representation used in bare JPEG files, but
1333 the surrounding PostScript code performs an inversion using the PS image
1334 operator. I am told that Photoshop 3.0 will write uninverted YCCK in
1335 EPS/JPEG files, and will omit the PS-level inversion. (But the data
1336 polarity used in bare JPEG files will not change in 3.0.) In either case,
1337 the JPEG library must not invert the data itself, or else Ghostscript would
1338 read these EPS files incorrectly.
1344 When the default error handler is used, any error detected inside the JPEG
1345 routines will cause a message to be printed on stderr, followed by exit().
1346 You can supply your own error handling routines to override this behavior
1347 and to control the treatment of nonfatal warnings and trace/debug messages.
1348 The file example.c illustrates the most common case, which is to have the
1349 application regain control after an error rather than exiting.
1351 The JPEG library never writes any message directly; it always goes through
1352 the error handling routines. Three classes of messages are recognized:
1353 * Fatal errors: the library cannot continue.
1354 * Warnings: the library can continue, but the data is corrupt, and a
1355 damaged output image is likely to result.
1356 * Trace/informational messages. These come with a trace level indicating
1357 the importance of the message; you can control the verbosity of the
1358 program by adjusting the maximum trace level that will be displayed.
1360 You may, if you wish, simply replace the entire JPEG error handling module
1361 (jerror.c) with your own code. However, you can avoid code duplication by
1362 only replacing some of the routines depending on the behavior you need.
1363 This is accomplished by calling jpeg_std_error() as usual, but then overriding
1364 some of the method pointers in the jpeg_error_mgr struct, as illustrated by
1367 All of the error handling routines will receive a pointer to the JPEG object
1368 (a j_common_ptr which points to either a jpeg_compress_struct or a
1369 jpeg_decompress_struct; if you need to tell which, test the is_decompressor
1370 field). This struct includes a pointer to the error manager struct in its
1371 "err" field. Frequently, custom error handler routines will need to access
1372 additional data which is not known to the JPEG library or the standard error
1373 handler. The most convenient way to do this is to embed either the JPEG
1374 object or the jpeg_error_mgr struct in a larger structure that contains
1375 additional fields; then casting the passed pointer provides access to the
1376 additional fields. Again, see example.c for one way to do it. (Beginning
1377 with IJG version 6b, there is also a void pointer "client_data" in each
1378 JPEG object, which the application can also use to find related data.
1379 The library does not touch client_data at all.)
1381 The individual methods that you might wish to override are:
1383 error_exit (j_common_ptr cinfo)
1384 Receives control for a fatal error. Information sufficient to
1385 generate the error message has been stored in cinfo->err; call
1386 output_message to display it. Control must NOT return to the caller;
1387 generally this routine will exit() or longjmp() somewhere.
1388 Typically you would override this routine to get rid of the exit()
1389 default behavior. Note that if you continue processing, you should
1390 clean up the JPEG object with jpeg_abort() or jpeg_destroy().
1392 output_message (j_common_ptr cinfo)
1393 Actual output of any JPEG message. Override this to send messages
1394 somewhere other than stderr. Note that this method does not know
1395 how to generate a message, only where to send it.
1397 format_message (j_common_ptr cinfo, char * buffer)
1398 Constructs a readable error message string based on the error info
1399 stored in cinfo->err. This method is called by output_message. Few
1400 applications should need to override this method. One possible
1401 reason for doing so is to implement dynamic switching of error message
1404 emit_message (j_common_ptr cinfo, int msg_level)
1405 Decide whether or not to emit a warning or trace message; if so,
1406 calls output_message. The main reason for overriding this method
1407 would be to abort on warnings. msg_level is -1 for warnings,
1408 0 and up for trace messages.
1410 Only error_exit() and emit_message() are called from the rest of the JPEG
1411 library; the other two are internal to the error handler.
1413 The actual message texts are stored in an array of strings which is pointed to
1414 by the field err->jpeg_message_table. The messages are numbered from 0 to
1415 err->last_jpeg_message, and it is these code numbers that are used in the
1416 JPEG library code. You could replace the message texts (for instance, with
1417 messages in French or German) by changing the message table pointer. See
1418 jerror.h for the default texts. CAUTION: this table will almost certainly
1419 change or grow from one library version to the next.
1421 It may be useful for an application to add its own message texts that are
1422 handled by the same mechanism. The error handler supports a second "add-on"
1423 message table for this purpose. To define an addon table, set the pointer
1424 err->addon_message_table and the message numbers err->first_addon_message and
1425 err->last_addon_message. If you number the addon messages beginning at 1000
1426 or so, you won't have to worry about conflicts with the library's built-in
1427 messages. See the sample applications cjpeg/djpeg for an example of using
1428 addon messages (the addon messages are defined in cderror.h).
1430 Actual invocation of the error handler is done via macros defined in jerror.h:
1431 ERREXITn(...) for fatal errors
1432 WARNMSn(...) for corrupt-data warnings
1433 TRACEMSn(...) for trace and informational messages.
1434 These macros store the message code and any additional parameters into the
1435 error handler struct, then invoke the error_exit() or emit_message() method.
1436 The variants of each macro are for varying numbers of additional parameters.
1437 The additional parameters are inserted into the generated message using
1438 standard printf() format codes.
1440 See jerror.h and jerror.c for further details.
1443 Compressed data handling (source and destination managers)
1444 ----------------------------------------------------------
1446 The JPEG compression library sends its compressed data to a "destination
1447 manager" module. The default destination manager just writes the data to a
1448 memory buffer or to a stdio stream, but you can provide your own manager to
1449 do something else. Similarly, the decompression library calls a "source
1450 manager" to obtain the compressed data; you can provide your own source
1451 manager if you want the data to come from somewhere other than a memory
1452 buffer or a stdio stream.
1454 In both cases, compressed data is processed a bufferload at a time: the
1455 destination or source manager provides a work buffer, and the library invokes
1456 the manager only when the buffer is filled or emptied. (You could define a
1457 one-character buffer to force the manager to be invoked for each byte, but
1458 that would be rather inefficient.) The buffer's size and location are
1459 controlled by the manager, not by the library. For example, the memory
1460 source manager just makes the buffer pointer and length point to the original
1461 data in memory. In this case the buffer-reload procedure will be invoked
1462 only if the decompressor ran off the end of the datastream, which would
1463 indicate an erroneous datastream.
1465 The work buffer is defined as an array of datatype JOCTET, which is generally
1466 "char" or "unsigned char". On a machine where char is not exactly 8 bits
1467 wide, you must define JOCTET as a wider data type and then modify the data
1468 source and destination modules to transcribe the work arrays into 8-bit units
1469 on external storage.
1471 A data destination manager struct contains a pointer and count defining the
1472 next byte to write in the work buffer and the remaining free space:
1474 JOCTET * next_output_byte; /* => next byte to write in buffer */
1475 size_t free_in_buffer; /* # of byte spaces remaining in buffer */
1477 The library increments the pointer and decrements the count until the buffer
1478 is filled. The manager's empty_output_buffer method must reset the pointer
1479 and count. The manager is expected to remember the buffer's starting address
1480 and total size in private fields not visible to the library.
1482 A data destination manager provides three methods:
1484 init_destination (j_compress_ptr cinfo)
1485 Initialize destination. This is called by jpeg_start_compress()
1486 before any data is actually written. It must initialize
1487 next_output_byte and free_in_buffer. free_in_buffer must be
1488 initialized to a positive value.
1490 empty_output_buffer (j_compress_ptr cinfo)
1491 This is called whenever the buffer has filled (free_in_buffer
1492 reaches zero). In typical applications, it should write out the
1493 *entire* buffer (use the saved start address and buffer length;
1494 ignore the current state of next_output_byte and free_in_buffer).
1495 Then reset the pointer & count to the start of the buffer, and
1496 return TRUE indicating that the buffer has been dumped.
1497 free_in_buffer must be set to a positive value when TRUE is
1498 returned. A FALSE return should only be used when I/O suspension is
1499 desired (this operating mode is discussed in the next section).
1501 term_destination (j_compress_ptr cinfo)
1502 Terminate destination --- called by jpeg_finish_compress() after all
1503 data has been written. In most applications, this must flush any
1504 data remaining in the buffer. Use either next_output_byte or
1505 free_in_buffer to determine how much data is in the buffer.
1507 term_destination() is NOT called by jpeg_abort() or jpeg_destroy(). If you
1508 want the destination manager to be cleaned up during an abort, you must do it
1511 You will also need code to create a jpeg_destination_mgr struct, fill in its
1512 method pointers, and insert a pointer to the struct into the "dest" field of
1513 the JPEG compression object. This can be done in-line in your setup code if
1514 you like, but it's probably cleaner to provide a separate routine similar to
1515 the jpeg_stdio_dest() or jpeg_mem_dest() routines of the supplied destination
1518 Decompression source managers follow a parallel design, but with some
1519 additional frammishes. The source manager struct contains a pointer and count
1520 defining the next byte to read from the work buffer and the number of bytes
1523 const JOCTET * next_input_byte; /* => next byte to read from buffer */
1524 size_t bytes_in_buffer; /* # of bytes remaining in buffer */
1526 The library increments the pointer and decrements the count until the buffer
1527 is emptied. The manager's fill_input_buffer method must reset the pointer and
1528 count. In most applications, the manager must remember the buffer's starting
1529 address and total size in private fields not visible to the library.
1531 A data source manager provides five methods:
1533 init_source (j_decompress_ptr cinfo)
1534 Initialize source. This is called by jpeg_read_header() before any
1535 data is actually read. Unlike init_destination(), it may leave
1536 bytes_in_buffer set to 0 (in which case a fill_input_buffer() call
1537 will occur immediately).
1539 fill_input_buffer (j_decompress_ptr cinfo)
1540 This is called whenever bytes_in_buffer has reached zero and more
1541 data is wanted. In typical applications, it should read fresh data
1542 into the buffer (ignoring the current state of next_input_byte and
1543 bytes_in_buffer), reset the pointer & count to the start of the
1544 buffer, and return TRUE indicating that the buffer has been reloaded.
1545 It is not necessary to fill the buffer entirely, only to obtain at
1546 least one more byte. bytes_in_buffer MUST be set to a positive value
1547 if TRUE is returned. A FALSE return should only be used when I/O
1548 suspension is desired (this mode is discussed in the next section).
1550 skip_input_data (j_decompress_ptr cinfo, long num_bytes)
1551 Skip num_bytes worth of data. The buffer pointer and count should
1552 be advanced over num_bytes input bytes, refilling the buffer as
1553 needed. This is used to skip over a potentially large amount of
1554 uninteresting data (such as an APPn marker). In some applications
1555 it may be possible to optimize away the reading of the skipped data,
1556 but it's not clear that being smart is worth much trouble; large
1557 skips are uncommon. bytes_in_buffer may be zero on return.
1558 A zero or negative skip count should be treated as a no-op.
1560 resync_to_restart (j_decompress_ptr cinfo, int desired)
1561 This routine is called only when the decompressor has failed to find
1562 a restart (RSTn) marker where one is expected. Its mission is to
1563 find a suitable point for resuming decompression. For most
1564 applications, we recommend that you just use the default resync
1565 procedure, jpeg_resync_to_restart(). However, if you are able to back
1566 up in the input data stream, or if you have a-priori knowledge about
1567 the likely location of restart markers, you may be able to do better.
1568 Read the read_restart_marker() and jpeg_resync_to_restart() routines
1569 in jdmarker.c if you think you'd like to implement your own resync
1572 term_source (j_decompress_ptr cinfo)
1573 Terminate source --- called by jpeg_finish_decompress() after all
1574 data has been read. Often a no-op.
1576 For both fill_input_buffer() and skip_input_data(), there is no such thing
1577 as an EOF return. If the end of the file has been reached, the routine has
1578 a choice of exiting via ERREXIT() or inserting fake data into the buffer.
1579 In most cases, generating a warning message and inserting a fake EOI marker
1580 is the best course of action --- this will allow the decompressor to output
1581 however much of the image is there. In pathological cases, the decompressor
1582 may swallow the EOI and again demand data ... just keep feeding it fake EOIs.
1583 jdatasrc.c illustrates the recommended error recovery behavior.
1585 term_source() is NOT called by jpeg_abort() or jpeg_destroy(). If you want
1586 the source manager to be cleaned up during an abort, you must do it yourself.
1588 You will also need code to create a jpeg_source_mgr struct, fill in its method
1589 pointers, and insert a pointer to the struct into the "src" field of the JPEG
1590 decompression object. This can be done in-line in your setup code if you
1591 like, but it's probably cleaner to provide a separate routine similar to the
1592 jpeg_stdio_src() or jpeg_mem_src() routines of the supplied source managers.
1594 For more information, consult the memory and stdio source and destination
1595 managers in jdatasrc.c and jdatadst.c.
1601 Some applications need to use the JPEG library as an incremental memory-to-
1602 memory filter: when the compressed data buffer is filled or emptied, they want
1603 control to return to the outer loop, rather than expecting that the buffer can
1604 be emptied or reloaded within the data source/destination manager subroutine.
1605 The library supports this need by providing an "I/O suspension" mode, which we
1606 describe in this section.
1608 The I/O suspension mode is not a panacea: nothing is guaranteed about the
1609 maximum amount of time spent in any one call to the library, so it will not
1610 eliminate response-time problems in single-threaded applications. If you
1611 need guaranteed response time, we suggest you "bite the bullet" and implement
1612 a real multi-tasking capability.
1614 To use I/O suspension, cooperation is needed between the calling application
1615 and the data source or destination manager; you will always need a custom
1616 source/destination manager. (Please read the previous section if you haven't
1617 already.) The basic idea is that the empty_output_buffer() or
1618 fill_input_buffer() routine is a no-op, merely returning FALSE to indicate
1619 that it has done nothing. Upon seeing this, the JPEG library suspends
1620 operation and returns to its caller. The surrounding application is
1621 responsible for emptying or refilling the work buffer before calling the
1624 Compression suspension:
1626 For compression suspension, use an empty_output_buffer() routine that returns
1627 FALSE; typically it will not do anything else. This will cause the
1628 compressor to return to the caller of jpeg_write_scanlines(), with the return
1629 value indicating that not all the supplied scanlines have been accepted.
1630 The application must make more room in the output buffer, adjust the output
1631 buffer pointer/count appropriately, and then call jpeg_write_scanlines()
1632 again, pointing to the first unconsumed scanline.
1634 When forced to suspend, the compressor will backtrack to a convenient stopping
1635 point (usually the start of the current MCU); it will regenerate some output
1636 data when restarted. Therefore, although empty_output_buffer() is only
1637 called when the buffer is filled, you should NOT write out the entire buffer
1638 after a suspension. Write only the data up to the current position of
1639 next_output_byte/free_in_buffer. The data beyond that point will be
1640 regenerated after resumption.
1642 Because of the backtracking behavior, a good-size output buffer is essential
1643 for efficiency; you don't want the compressor to suspend often. (In fact, an
1644 overly small buffer could lead to infinite looping, if a single MCU required
1645 more data than would fit in the buffer.) We recommend a buffer of at least
1646 several Kbytes. You may want to insert explicit code to ensure that you don't
1647 call jpeg_write_scanlines() unless there is a reasonable amount of space in
1648 the output buffer; in other words, flush the buffer before trying to compress
1651 The compressor does not allow suspension while it is trying to write JPEG
1652 markers at the beginning and end of the file. This means that:
1653 * At the beginning of a compression operation, there must be enough free
1654 space in the output buffer to hold the header markers (typically 600 or
1655 so bytes). The recommended buffer size is bigger than this anyway, so
1656 this is not a problem as long as you start with an empty buffer. However,
1657 this restriction might catch you if you insert large special markers, such
1658 as a JFIF thumbnail image, without flushing the buffer afterwards.
1659 * When you call jpeg_finish_compress(), there must be enough space in the
1660 output buffer to emit any buffered data and the final EOI marker. In the
1661 current implementation, half a dozen bytes should suffice for this, but
1662 for safety's sake we recommend ensuring that at least 100 bytes are free
1663 before calling jpeg_finish_compress().
1665 A more significant restriction is that jpeg_finish_compress() cannot suspend.
1666 This means you cannot use suspension with multi-pass operating modes, namely
1667 Huffman code optimization and multiple-scan output. Those modes write the
1668 whole file during jpeg_finish_compress(), which will certainly result in
1669 buffer overrun. (Note that this restriction applies only to compression,
1670 not decompression. The decompressor supports input suspension in all of its
1673 Decompression suspension:
1675 For decompression suspension, use a fill_input_buffer() routine that simply
1676 returns FALSE (except perhaps during error recovery, as discussed below).
1677 This will cause the decompressor to return to its caller with an indication
1678 that suspension has occurred. This can happen at four places:
1679 * jpeg_read_header(): will return JPEG_SUSPENDED.
1680 * jpeg_start_decompress(): will return FALSE, rather than its usual TRUE.
1681 * jpeg_read_scanlines(): will return the number of scanlines already
1682 completed (possibly 0).
1683 * jpeg_finish_decompress(): will return FALSE, rather than its usual TRUE.
1684 The surrounding application must recognize these cases, load more data into
1685 the input buffer, and repeat the call. In the case of jpeg_read_scanlines(),
1686 increment the passed pointers past any scanlines successfully read.
1688 Just as with compression, the decompressor will typically backtrack to a
1689 convenient restart point before suspending. When fill_input_buffer() is
1690 called, next_input_byte/bytes_in_buffer point to the current restart point,
1691 which is where the decompressor will backtrack to if FALSE is returned.
1692 The data beyond that position must NOT be discarded if you suspend; it needs
1693 to be re-read upon resumption. In most implementations, you'll need to shift
1694 this data down to the start of your work buffer and then load more data after
1695 it. Again, this behavior means that a several-Kbyte work buffer is essential
1696 for decent performance; furthermore, you should load a reasonable amount of
1697 new data before resuming decompression. (If you loaded, say, only one new
1698 byte each time around, you could waste a LOT of cycles.)
1700 The skip_input_data() source manager routine requires special care in a
1701 suspension scenario. This routine is NOT granted the ability to suspend the
1702 decompressor; it can decrement bytes_in_buffer to zero, but no more. If the
1703 requested skip distance exceeds the amount of data currently in the input
1704 buffer, then skip_input_data() must set bytes_in_buffer to zero and record the
1705 additional skip distance somewhere else. The decompressor will immediately
1706 call fill_input_buffer(), which should return FALSE, which will cause a
1707 suspension return. The surrounding application must then arrange to discard
1708 the recorded number of bytes before it resumes loading the input buffer.
1709 (Yes, this design is rather baroque, but it avoids complexity in the far more
1710 common case where a non-suspending source manager is used.)
1712 If the input data has been exhausted, we recommend that you emit a warning
1713 and insert dummy EOI markers just as a non-suspending data source manager
1714 would do. This can be handled either in the surrounding application logic or
1715 within fill_input_buffer(); the latter is probably more efficient. If
1716 fill_input_buffer() knows that no more data is available, it can set the
1717 pointer/count to point to a dummy EOI marker and then return TRUE just as
1718 though it had read more data in a non-suspending situation.
1720 The decompressor does not attempt to suspend within standard JPEG markers;
1721 instead it will backtrack to the start of the marker and reprocess the whole
1722 marker next time. Hence the input buffer must be large enough to hold the
1723 longest standard marker in the file. Standard JPEG markers should normally
1724 not exceed a few hundred bytes each (DHT tables are typically the longest).
1725 We recommend at least a 2K buffer for performance reasons, which is much
1726 larger than any correct marker is likely to be. For robustness against
1727 damaged marker length counts, you may wish to insert a test in your
1728 application for the case that the input buffer is completely full and yet
1729 the decoder has suspended without consuming any data --- otherwise, if this
1730 situation did occur, it would lead to an endless loop. (The library can't
1731 provide this test since it has no idea whether "the buffer is full", or
1732 even whether there is a fixed-size input buffer.)
1734 The input buffer would need to be 64K to allow for arbitrary COM or APPn
1735 markers, but these are handled specially: they are either saved into allocated
1736 memory, or skipped over by calling skip_input_data(). In the former case,
1737 suspension is handled correctly, and in the latter case, the problem of
1738 buffer overrun is placed on skip_input_data's shoulders, as explained above.
1739 Note that if you provide your own marker handling routine for large markers,
1740 you should consider how to deal with buffer overflow.
1742 Multiple-buffer management:
1744 In some applications it is desirable to store the compressed data in a linked
1745 list of buffer areas, so as to avoid data copying. This can be handled by
1746 having empty_output_buffer() or fill_input_buffer() set the pointer and count
1747 to reference the next available buffer; FALSE is returned only if no more
1748 buffers are available. Although seemingly straightforward, there is a
1749 pitfall in this approach: the backtrack that occurs when FALSE is returned
1750 could back up into an earlier buffer. For example, when fill_input_buffer()
1751 is called, the current pointer & count indicate the backtrack restart point.
1752 Since fill_input_buffer() will set the pointer and count to refer to a new
1753 buffer, the restart position must be saved somewhere else. Suppose a second
1754 call to fill_input_buffer() occurs in the same library call, and no
1755 additional input data is available, so fill_input_buffer must return FALSE.
1756 If the JPEG library has not moved the pointer/count forward in the current
1757 buffer, then *the correct restart point is the saved position in the prior
1758 buffer*. Prior buffers may be discarded only after the library establishes
1759 a restart point within a later buffer. Similar remarks apply for output into
1762 The library will never attempt to backtrack over a skip_input_data() call,
1763 so any skipped data can be permanently discarded. You still have to deal
1764 with the case of skipping not-yet-received data, however.
1766 It's much simpler to use only a single buffer; when fill_input_buffer() is
1767 called, move any unconsumed data (beyond the current pointer/count) down to
1768 the beginning of this buffer and then load new data into the remaining buffer
1769 space. This approach requires a little more data copying but is far easier
1773 Progressive JPEG support
1774 ------------------------
1776 Progressive JPEG rearranges the stored data into a series of scans of
1777 increasing quality. In situations where a JPEG file is transmitted across a
1778 slow communications link, a decoder can generate a low-quality image very
1779 quickly from the first scan, then gradually improve the displayed quality as
1780 more scans are received. The final image after all scans are complete is
1781 identical to that of a regular (sequential) JPEG file of the same quality
1782 setting. Progressive JPEG files are often slightly smaller than equivalent
1783 sequential JPEG files, but the possibility of incremental display is the main
1784 reason for using progressive JPEG.
1786 The IJG encoder library generates progressive JPEG files when given a
1787 suitable "scan script" defining how to divide the data into scans.
1788 Creation of progressive JPEG files is otherwise transparent to the encoder.
1789 Progressive JPEG files can also be read transparently by the decoder library.
1790 If the decoding application simply uses the library as defined above, it
1791 will receive a final decoded image without any indication that the file was
1792 progressive. Of course, this approach does not allow incremental display.
1793 To perform incremental display, an application needs to use the decoder
1794 library's "buffered-image" mode, in which it receives a decoded image
1797 Each displayed scan requires about as much work to decode as a full JPEG
1798 image of the same size, so the decoder must be fairly fast in relation to the
1799 data transmission rate in order to make incremental display useful. However,
1800 it is possible to skip displaying the image and simply add the incoming bits
1801 to the decoder's coefficient buffer. This is fast because only Huffman
1802 decoding need be done, not IDCT, upsampling, colorspace conversion, etc.
1803 The IJG decoder library allows the application to switch dynamically between
1804 displaying the image and simply absorbing the incoming bits. A properly
1805 coded application can automatically adapt the number of display passes to
1806 suit the time available as the image is received. Also, a final
1807 higher-quality display cycle can be performed from the buffered data after
1808 the end of the file is reached.
1810 Progressive compression:
1812 To create a progressive JPEG file (or a multiple-scan sequential JPEG file),
1813 set the scan_info cinfo field to point to an array of scan descriptors, and
1814 perform compression as usual. Instead of constructing your own scan list,
1815 you can call the jpeg_simple_progression() helper routine to create a
1816 recommended progression sequence; this method should be used by all
1817 applications that don't want to get involved in the nitty-gritty of
1818 progressive scan sequence design. (If you want to provide user control of
1819 scan sequences, you may wish to borrow the scan script reading code found
1820 in rdswitch.c, so that you can read scan script files just like cjpeg's.)
1821 When scan_info is not NULL, the compression library will store DCT'd data
1822 into a buffer array as jpeg_write_scanlines() is called, and will emit all
1823 the requested scans during jpeg_finish_compress(). This implies that
1824 multiple-scan output cannot be created with a suspending data destination
1825 manager, since jpeg_finish_compress() does not support suspension. We
1826 should also note that the compressor currently forces Huffman optimization
1827 mode when creating a progressive JPEG file, because the default Huffman
1828 tables are unsuitable for progressive files.
1830 Progressive decompression:
1832 When buffered-image mode is not used, the decoder library will read all of
1833 a multi-scan file during jpeg_start_decompress(), so that it can provide a
1834 final decoded image. (Here "multi-scan" means either progressive or
1835 multi-scan sequential.) This makes multi-scan files transparent to the
1836 decoding application. However, existing applications that used suspending
1837 input with version 5 of the IJG library will need to be modified to check
1838 for a suspension return from jpeg_start_decompress().
1840 To perform incremental display, an application must use the library's
1841 buffered-image mode. This is described in the next section.
1847 In buffered-image mode, the library stores the partially decoded image in a
1848 coefficient buffer, from which it can be read out as many times as desired.
1849 This mode is typically used for incremental display of progressive JPEG files,
1850 but it can be used with any JPEG file. Each scan of a progressive JPEG file
1851 adds more data (more detail) to the buffered image. The application can
1852 display in lockstep with the source file (one display pass per input scan),
1853 or it can allow input processing to outrun display processing. By making
1854 input and display processing run independently, it is possible for the
1855 application to adapt progressive display to a wide range of data transmission
1858 The basic control flow for buffered-image decoding is
1860 jpeg_create_decompress()
1863 set overall decompression parameters
1864 cinfo.buffered_image = TRUE; /* select buffered-image mode */
1865 jpeg_start_decompress()
1866 for (each output pass) {
1867 adjust output decompression parameters if required
1868 jpeg_start_output() /* start a new output pass */
1869 for (all scanlines in image) {
1870 jpeg_read_scanlines()
1873 jpeg_finish_output() /* terminate output pass */
1875 jpeg_finish_decompress()
1876 jpeg_destroy_decompress()
1878 This differs from ordinary unbuffered decoding in that there is an additional
1879 level of looping. The application can choose how many output passes to make
1880 and how to display each pass.
1882 The simplest approach to displaying progressive images is to do one display
1883 pass for each scan appearing in the input file. In this case the outer loop
1884 condition is typically
1885 while (! jpeg_input_complete(&cinfo))
1886 and the start-output call should read
1887 jpeg_start_output(&cinfo, cinfo.input_scan_number);
1888 The second parameter to jpeg_start_output() indicates which scan of the input
1889 file is to be displayed; the scans are numbered starting at 1 for this
1890 purpose. (You can use a loop counter starting at 1 if you like, but using
1891 the library's input scan counter is easier.) The library automatically reads
1892 data as necessary to complete each requested scan, and jpeg_finish_output()
1893 advances to the next scan or end-of-image marker (hence input_scan_number
1894 will be incremented by the time control arrives back at jpeg_start_output()).
1895 With this technique, data is read from the input file only as needed, and
1896 input and output processing run in lockstep.
1898 After reading the final scan and reaching the end of the input file, the
1899 buffered image remains available; it can be read additional times by
1900 repeating the jpeg_start_output()/jpeg_read_scanlines()/jpeg_finish_output()
1901 sequence. For example, a useful technique is to use fast one-pass color
1902 quantization for display passes made while the image is arriving, followed by
1903 a final display pass using two-pass quantization for highest quality. This
1904 is done by changing the library parameters before the final output pass.
1905 Changing parameters between passes is discussed in detail below.
1907 In general the last scan of a progressive file cannot be recognized as such
1908 until after it is read, so a post-input display pass is the best approach if
1909 you want special processing in the final pass.
1911 When done with the image, be sure to call jpeg_finish_decompress() to release
1912 the buffered image (or just use jpeg_destroy_decompress()).
1914 If input data arrives faster than it can be displayed, the application can
1915 cause the library to decode input data in advance of what's needed to produce
1916 output. This is done by calling the routine jpeg_consume_input().
1917 The return value is one of the following:
1918 JPEG_REACHED_SOS: reached an SOS marker (the start of a new scan)
1919 JPEG_REACHED_EOI: reached the EOI marker (end of image)
1920 JPEG_ROW_COMPLETED: completed reading one MCU row of compressed data
1921 JPEG_SCAN_COMPLETED: completed reading last MCU row of current scan
1922 JPEG_SUSPENDED: suspended before completing any of the above
1923 (JPEG_SUSPENDED can occur only if a suspending data source is used.) This
1924 routine can be called at any time after initializing the JPEG object. It
1925 reads some additional data and returns when one of the indicated significant
1926 events occurs. (If called after the EOI marker is reached, it will
1927 immediately return JPEG_REACHED_EOI without attempting to read more data.)
1929 The library's output processing will automatically call jpeg_consume_input()
1930 whenever the output processing overtakes the input; thus, simple lockstep
1931 display requires no direct calls to jpeg_consume_input(). But by adding
1932 calls to jpeg_consume_input(), you can absorb data in advance of what is
1933 being displayed. This has two benefits:
1934 * You can limit buildup of unprocessed data in your input buffer.
1935 * You can eliminate extra display passes by paying attention to the
1936 state of the library's input processing.
1938 The first of these benefits only requires interspersing calls to
1939 jpeg_consume_input() with your display operations and any other processing
1940 you may be doing. To avoid wasting cycles due to backtracking, it's best to
1941 call jpeg_consume_input() only after a hundred or so new bytes have arrived.
1942 This is discussed further under "I/O suspension", above. (Note: the JPEG
1943 library currently is not thread-safe. You must not call jpeg_consume_input()
1944 from one thread of control if a different library routine is working on the
1945 same JPEG object in another thread.)
1947 When input arrives fast enough that more than one new scan is available
1948 before you start a new output pass, you may as well skip the output pass
1949 corresponding to the completed scan. This occurs for free if you pass
1950 cinfo.input_scan_number as the target scan number to jpeg_start_output().
1951 The input_scan_number field is simply the index of the scan currently being
1952 consumed by the input processor. You can ensure that this is up-to-date by
1953 emptying the input buffer just before calling jpeg_start_output(): call
1954 jpeg_consume_input() repeatedly until it returns JPEG_SUSPENDED or
1957 The target scan number passed to jpeg_start_output() is saved in the
1958 cinfo.output_scan_number field. The library's output processing calls
1959 jpeg_consume_input() whenever the current input scan number and row within
1960 that scan is less than or equal to the current output scan number and row.
1961 Thus, input processing can "get ahead" of the output processing but is not
1962 allowed to "fall behind". You can achieve several different effects by
1963 manipulating this interlock rule. For example, if you pass a target scan
1964 number greater than the current input scan number, the output processor will
1965 wait until that scan starts to arrive before producing any output. (To avoid
1966 an infinite loop, the target scan number is automatically reset to the last
1967 scan number when the end of image is reached. Thus, if you specify a large
1968 target scan number, the library will just absorb the entire input file and
1969 then perform an output pass. This is effectively the same as what
1970 jpeg_start_decompress() does when you don't select buffered-image mode.)
1971 When you pass a target scan number equal to the current input scan number,
1972 the image is displayed no faster than the current input scan arrives. The
1973 final possibility is to pass a target scan number less than the current input
1974 scan number; this disables the input/output interlock and causes the output
1975 processor to simply display whatever it finds in the image buffer, without
1976 waiting for input. (However, the library will not accept a target scan
1977 number less than one, so you can't avoid waiting for the first scan.)
1979 When data is arriving faster than the output display processing can advance
1980 through the image, jpeg_consume_input() will store data into the buffered
1981 image beyond the point at which the output processing is reading data out
1982 again. If the input arrives fast enough, it may "wrap around" the buffer to
1983 the point where the input is more than one whole scan ahead of the output.
1984 If the output processing simply proceeds through its display pass without
1985 paying attention to the input, the effect seen on-screen is that the lower
1986 part of the image is one or more scans better in quality than the upper part.
1987 Then, when the next output scan is started, you have a choice of what target
1988 scan number to use. The recommended choice is to use the current input scan
1989 number at that time, which implies that you've skipped the output scans
1990 corresponding to the input scans that were completed while you processed the
1991 previous output scan. In this way, the decoder automatically adapts its
1992 speed to the arriving data, by skipping output scans as necessary to keep up
1993 with the arriving data.
1995 When using this strategy, you'll want to be sure that you perform a final
1996 output pass after receiving all the data; otherwise your last display may not
1997 be full quality across the whole screen. So the right outer loop logic is
1998 something like this:
2000 absorb any waiting input by calling jpeg_consume_input()
2001 final_pass = jpeg_input_complete(&cinfo);
2002 adjust output decompression parameters if required
2003 jpeg_start_output(&cinfo, cinfo.input_scan_number);
2005 jpeg_finish_output()
2006 } while (! final_pass);
2007 rather than quitting as soon as jpeg_input_complete() returns TRUE. This
2008 arrangement makes it simple to use higher-quality decoding parameters
2009 for the final pass. But if you don't want to use special parameters for
2010 the final pass, the right loop logic is like this:
2012 absorb any waiting input by calling jpeg_consume_input()
2013 jpeg_start_output(&cinfo, cinfo.input_scan_number);
2015 jpeg_finish_output()
2016 if (jpeg_input_complete(&cinfo) &&
2017 cinfo.input_scan_number == cinfo.output_scan_number)
2020 In this case you don't need to know in advance whether an output pass is to
2021 be the last one, so it's not necessary to have reached EOF before starting
2022 the final output pass; rather, what you want to test is whether the output
2023 pass was performed in sync with the final input scan. This form of the loop
2024 will avoid an extra output pass whenever the decoder is able (or nearly able)
2025 to keep up with the incoming data.
2027 When the data transmission speed is high, you might begin a display pass,
2028 then find that much or all of the file has arrived before you can complete
2029 the pass. (You can detect this by noting the JPEG_REACHED_EOI return code
2030 from jpeg_consume_input(), or equivalently by testing jpeg_input_complete().)
2031 In this situation you may wish to abort the current display pass and start a
2032 new one using the newly arrived information. To do so, just call
2033 jpeg_finish_output() and then start a new pass with jpeg_start_output().
2035 A variant strategy is to abort and restart display if more than one complete
2036 scan arrives during an output pass; this can be detected by noting
2037 JPEG_REACHED_SOS returns and/or examining cinfo.input_scan_number. This
2038 idea should be employed with caution, however, since the display process
2039 might never get to the bottom of the image before being aborted, resulting
2040 in the lower part of the screen being several passes worse than the upper.
2041 In most cases it's probably best to abort an output pass only if the whole
2042 file has arrived and you want to begin the final output pass immediately.
2044 When receiving data across a communication link, we recommend always using
2045 the current input scan number for the output target scan number; if a
2046 higher-quality final pass is to be done, it should be started (aborting any
2047 incomplete output pass) as soon as the end of file is received. However,
2048 many other strategies are possible. For example, the application can examine
2049 the parameters of the current input scan and decide whether to display it or
2050 not. If the scan contains only chroma data, one might choose not to use it
2051 as the target scan, expecting that the scan will be small and will arrive
2052 quickly. To skip to the next scan, call jpeg_consume_input() until it
2053 returns JPEG_REACHED_SOS or JPEG_REACHED_EOI. Or just use the next higher
2054 number as the target scan for jpeg_start_output(); but that method doesn't
2055 let you inspect the next scan's parameters before deciding to display it.
2058 In buffered-image mode, jpeg_start_decompress() never performs input and
2059 thus never suspends. An application that uses input suspension with
2060 buffered-image mode must be prepared for suspension returns from these
2062 * jpeg_start_output() performs input only if you request 2-pass quantization
2063 and the target scan isn't fully read yet. (This is discussed below.)
2064 * jpeg_read_scanlines(), as always, returns the number of scanlines that it
2065 was able to produce before suspending.
2066 * jpeg_finish_output() will read any markers following the target scan,
2067 up to the end of the file or the SOS marker that begins another scan.
2068 (But it reads no input if jpeg_consume_input() has already reached the
2069 end of the file or a SOS marker beyond the target output scan.)
2070 * jpeg_finish_decompress() will read until the end of file, and thus can
2071 suspend if the end hasn't already been reached (as can be tested by
2072 calling jpeg_input_complete()).
2073 jpeg_start_output(), jpeg_finish_output(), and jpeg_finish_decompress()
2074 all return TRUE if they completed their tasks, FALSE if they had to suspend.
2075 In the event of a FALSE return, the application must load more input data
2076 and repeat the call. Applications that use non-suspending data sources need
2077 not check the return values of these three routines.
2080 It is possible to change decoding parameters between output passes in the
2081 buffered-image mode. The decoder library currently supports only very
2082 limited changes of parameters. ONLY THE FOLLOWING parameter changes are
2083 allowed after jpeg_start_decompress() is called:
2084 * dct_method can be changed before each call to jpeg_start_output().
2085 For example, one could use a fast DCT method for early scans, changing
2086 to a higher quality method for the final scan.
2087 * dither_mode can be changed before each call to jpeg_start_output();
2088 of course this has no impact if not using color quantization. Typically
2089 one would use ordered dither for initial passes, then switch to
2090 Floyd-Steinberg dither for the final pass. Caution: changing dither mode
2091 can cause more memory to be allocated by the library. Although the amount
2092 of memory involved is not large (a scanline or so), it may cause the
2093 initial max_memory_to_use specification to be exceeded, which in the worst
2094 case would result in an out-of-memory failure.
2095 * do_block_smoothing can be changed before each call to jpeg_start_output().
2096 This setting is relevant only when decoding a progressive JPEG image.
2097 During the first DC-only scan, block smoothing provides a very "fuzzy" look
2098 instead of the very "blocky" look seen without it; which is better seems a
2099 matter of personal taste. But block smoothing is nearly always a win
2100 during later stages, especially when decoding a successive-approximation
2101 image: smoothing helps to hide the slight blockiness that otherwise shows
2102 up on smooth gradients until the lowest coefficient bits are sent.
2103 * Color quantization mode can be changed under the rules described below.
2104 You *cannot* change between full-color and quantized output (because that
2105 would alter the required I/O buffer sizes), but you can change which
2106 quantization method is used.
2108 When generating color-quantized output, changing quantization method is a
2109 very useful way of switching between high-speed and high-quality display.
2110 The library allows you to change among its three quantization methods:
2111 1. Single-pass quantization to a fixed color cube.
2112 Selected by cinfo.two_pass_quantize = FALSE and cinfo.colormap = NULL.
2113 2. Single-pass quantization to an application-supplied colormap.
2114 Selected by setting cinfo.colormap to point to the colormap (the value of
2115 two_pass_quantize is ignored); also set cinfo.actual_number_of_colors.
2116 3. Two-pass quantization to a colormap chosen specifically for the image.
2117 Selected by cinfo.two_pass_quantize = TRUE and cinfo.colormap = NULL.
2118 (This is the default setting selected by jpeg_read_header, but it is
2119 probably NOT what you want for the first pass of progressive display!)
2120 These methods offer successively better quality and lesser speed. However,
2121 only the first method is available for quantizing in non-RGB color spaces.
2123 IMPORTANT: because the different quantizer methods have very different
2124 working-storage requirements, the library requires you to indicate which
2125 one(s) you intend to use before you call jpeg_start_decompress(). (If we did
2126 not require this, the max_memory_to_use setting would be a complete fiction.)
2127 You do this by setting one or more of these three cinfo fields to TRUE:
2128 enable_1pass_quant Fixed color cube colormap
2129 enable_external_quant Externally-supplied colormap
2130 enable_2pass_quant Two-pass custom colormap
2131 All three are initialized FALSE by jpeg_read_header(). But
2132 jpeg_start_decompress() automatically sets TRUE the one selected by the
2133 current two_pass_quantize and colormap settings, so you only need to set the
2134 enable flags for any other quantization methods you plan to change to later.
2136 After setting the enable flags correctly at jpeg_start_decompress() time, you
2137 can change to any enabled quantization method by setting two_pass_quantize
2138 and colormap properly just before calling jpeg_start_output(). The following
2139 special rules apply:
2140 1. You must explicitly set cinfo.colormap to NULL when switching to 1-pass
2141 or 2-pass mode from a different mode, or when you want the 2-pass
2142 quantizer to be re-run to generate a new colormap.
2143 2. To switch to an external colormap, or to change to a different external
2144 colormap than was used on the prior pass, you must call
2145 jpeg_new_colormap() after setting cinfo.colormap.
2146 NOTE: if you want to use the same colormap as was used in the prior pass,
2147 you should not do either of these things. This will save some nontrivial
2149 (These requirements exist because cinfo.colormap will always be non-NULL
2150 after completing a prior output pass, since both the 1-pass and 2-pass
2151 quantizers set it to point to their output colormaps. Thus you have to
2152 do one of these two things to notify the library that something has changed.
2153 Yup, it's a bit klugy, but it's necessary to do it this way for backwards
2156 Note that in buffered-image mode, the library generates any requested colormap
2157 during jpeg_start_output(), not during jpeg_start_decompress().
2159 When using two-pass quantization, jpeg_start_output() makes a pass over the
2160 buffered image to determine the optimum color map; it therefore may take a
2161 significant amount of time, whereas ordinarily it does little work. The
2162 progress monitor hook is called during this pass, if defined. It is also
2163 important to realize that if the specified target scan number is greater than
2164 or equal to the current input scan number, jpeg_start_output() will attempt
2165 to consume input as it makes this pass. If you use a suspending data source,
2166 you need to check for a FALSE return from jpeg_start_output() under these
2167 conditions. The combination of 2-pass quantization and a not-yet-fully-read
2168 target scan is the only case in which jpeg_start_output() will consume input.
2171 Application authors who support buffered-image mode may be tempted to use it
2172 for all JPEG images, even single-scan ones. This will work, but it is
2173 inefficient: there is no need to create an image-sized coefficient buffer for
2174 single-scan images. Requesting buffered-image mode for such an image wastes
2175 memory. Worse, it can cost time on large images, since the buffered data has
2176 to be swapped out or written to a temporary file. If you are concerned about
2177 maximum performance on baseline JPEG files, you should use buffered-image
2178 mode only when the incoming file actually has multiple scans. This can be
2179 tested by calling jpeg_has_multiple_scans(), which will return a correct
2180 result at any time after jpeg_read_header() completes.
2182 It is also worth noting that when you use jpeg_consume_input() to let input
2183 processing get ahead of output processing, the resulting pattern of access to
2184 the coefficient buffer is quite nonsequential. It's best to use the memory
2185 manager jmemnobs.c if you can (ie, if you have enough real or virtual main
2186 memory). If not, at least make sure that max_memory_to_use is set as high as
2187 possible. If the JPEG memory manager has to use a temporary file, you will
2188 probably see a lot of disk traffic and poor performance. (This could be
2189 improved with additional work on the memory manager, but we haven't gotten
2192 In some applications it may be convenient to use jpeg_consume_input() for all
2193 input processing, including reading the initial markers; that is, you may
2194 wish to call jpeg_consume_input() instead of jpeg_read_header() during
2195 startup. This works, but note that you must check for JPEG_REACHED_SOS and
2196 JPEG_REACHED_EOI return codes as the equivalent of jpeg_read_header's codes.
2197 Once the first SOS marker has been reached, you must call
2198 jpeg_start_decompress() before jpeg_consume_input() will consume more input;
2199 it'll just keep returning JPEG_REACHED_SOS until you do. If you read a
2200 tables-only file this way, jpeg_consume_input() will return JPEG_REACHED_EOI
2201 without ever returning JPEG_REACHED_SOS; be sure to check for this case.
2202 If this happens, the decompressor will not read any more input until you call
2203 jpeg_abort() to reset it. It is OK to call jpeg_consume_input() even when not
2204 using buffered-image mode, but in that case it's basically a no-op after the
2205 initial markers have been read: it will just return JPEG_SUSPENDED.
2208 Abbreviated datastreams and multiple images
2209 -------------------------------------------
2211 A JPEG compression or decompression object can be reused to process multiple
2212 images. This saves a small amount of time per image by eliminating the
2213 "create" and "destroy" operations, but that isn't the real purpose of the
2214 feature. Rather, reuse of an object provides support for abbreviated JPEG
2215 datastreams. Object reuse can also simplify processing a series of images in
2216 a single input or output file. This section explains these features.
2218 A JPEG file normally contains several hundred bytes worth of quantization
2219 and Huffman tables. In a situation where many images will be stored or
2220 transmitted with identical tables, this may represent an annoying overhead.
2221 The JPEG standard therefore permits tables to be omitted. The standard
2222 defines three classes of JPEG datastreams:
2223 * "Interchange" datastreams contain an image and all tables needed to decode
2224 the image. These are the usual kind of JPEG file.
2225 * "Abbreviated image" datastreams contain an image, but are missing some or
2226 all of the tables needed to decode that image.
2227 * "Abbreviated table specification" (henceforth "tables-only") datastreams
2228 contain only table specifications.
2229 To decode an abbreviated image, it is necessary to load the missing table(s)
2230 into the decoder beforehand. This can be accomplished by reading a separate
2231 tables-only file. A variant scheme uses a series of images in which the first
2232 image is an interchange (complete) datastream, while subsequent ones are
2233 abbreviated and rely on the tables loaded by the first image. It is assumed
2234 that once the decoder has read a table, it will remember that table until a
2235 new definition for the same table number is encountered.
2237 It is the application designer's responsibility to figure out how to associate
2238 the correct tables with an abbreviated image. While abbreviated datastreams
2239 can be useful in a closed environment, their use is strongly discouraged in
2240 any situation where data exchange with other applications might be needed.
2243 The JPEG library provides support for reading and writing any combination of
2244 tables-only datastreams and abbreviated images. In both compression and
2245 decompression objects, a quantization or Huffman table will be retained for
2246 the lifetime of the object, unless it is overwritten by a new table definition.
2249 To create abbreviated image datastreams, it is only necessary to tell the
2250 compressor not to emit some or all of the tables it is using. Each
2251 quantization and Huffman table struct contains a boolean field "sent_table",
2252 which normally is initialized to FALSE. For each table used by the image, the
2253 header-writing process emits the table and sets sent_table = TRUE unless it is
2254 already TRUE. (In normal usage, this prevents outputting the same table
2255 definition multiple times, as would otherwise occur because the chroma
2256 components typically share tables.) Thus, setting this field to TRUE before
2257 calling jpeg_start_compress() will prevent the table from being written at
2260 If you want to create a "pure" abbreviated image file containing no tables,
2261 just call "jpeg_suppress_tables(&cinfo, TRUE)" after constructing all the
2262 tables. If you want to emit some but not all tables, you'll need to set the
2263 individual sent_table fields directly.
2265 To create an abbreviated image, you must also call jpeg_start_compress()
2266 with a second parameter of FALSE, not TRUE. Otherwise jpeg_start_compress()
2267 will force all the sent_table fields to FALSE. (This is a safety feature to
2268 prevent abbreviated images from being created accidentally.)
2270 To create a tables-only file, perform the same parameter setup that you
2271 normally would, but instead of calling jpeg_start_compress() and so on, call
2272 jpeg_write_tables(&cinfo). This will write an abbreviated datastream
2273 containing only SOI, DQT and/or DHT markers, and EOI. All the quantization
2274 and Huffman tables that are currently defined in the compression object will
2275 be emitted unless their sent_tables flag is already TRUE, and then all the
2276 sent_tables flags will be set TRUE.
2278 A sure-fire way to create matching tables-only and abbreviated image files
2279 is to proceed as follows:
2281 create JPEG compression object
2283 set destination to tables-only file
2284 jpeg_write_tables(&cinfo);
2285 set destination to image file
2286 jpeg_start_compress(&cinfo, FALSE);
2288 jpeg_finish_compress(&cinfo);
2290 Since the JPEG parameters are not altered between writing the table file and
2291 the abbreviated image file, the same tables are sure to be used. Of course,
2292 you can repeat the jpeg_start_compress() ... jpeg_finish_compress() sequence
2293 many times to produce many abbreviated image files matching the table file.
2295 You cannot suppress output of the computed Huffman tables when Huffman
2296 optimization is selected. (If you could, there'd be no way to decode the
2297 image...) Generally, you don't want to set optimize_coding = TRUE when
2298 you are trying to produce abbreviated files.
2300 In some cases you might want to compress an image using tables which are
2301 not stored in the application, but are defined in an interchange or
2302 tables-only file readable by the application. This can be done by setting up
2303 a JPEG decompression object to read the specification file, then copying the
2304 tables into your compression object. See jpeg_copy_critical_parameters()
2305 for an example of copying quantization tables.
2308 To read abbreviated image files, you simply need to load the proper tables
2309 into the decompression object before trying to read the abbreviated image.
2310 If the proper tables are stored in the application program, you can just
2311 allocate the table structs and fill in their contents directly. For example,
2312 to load a fixed quantization table into table slot "n":
2314 if (cinfo.quant_tbl_ptrs[n] == NULL)
2315 cinfo.quant_tbl_ptrs[n] = jpeg_alloc_quant_table((j_common_ptr) &cinfo);
2316 quant_ptr = cinfo.quant_tbl_ptrs[n]; /* quant_ptr is JQUANT_TBL* */
2317 for (i = 0; i < 64; i++) {
2318 /* Qtable[] is desired quantization table, in natural array order */
2319 quant_ptr->quantval[i] = Qtable[i];
2322 Code to load a fixed Huffman table is typically (for AC table "n"):
2324 if (cinfo.ac_huff_tbl_ptrs[n] == NULL)
2325 cinfo.ac_huff_tbl_ptrs[n] = jpeg_alloc_huff_table((j_common_ptr) &cinfo);
2326 huff_ptr = cinfo.ac_huff_tbl_ptrs[n]; /* huff_ptr is JHUFF_TBL* */
2327 for (i = 1; i <= 16; i++) {
2328 /* counts[i] is number of Huffman codes of length i bits, i=1..16 */
2329 huff_ptr->bits[i] = counts[i];
2331 for (i = 0; i < 256; i++) {
2332 /* symbols[] is the list of Huffman symbols, in code-length order */
2333 huff_ptr->huffval[i] = symbols[i];
2336 (Note that trying to set cinfo.quant_tbl_ptrs[n] to point directly at a
2337 constant JQUANT_TBL object is not safe. If the incoming file happened to
2338 contain a quantization table definition, your master table would get
2339 overwritten! Instead allocate a working table copy and copy the master table
2340 into it, as illustrated above. Ditto for Huffman tables, of course.)
2342 You might want to read the tables from a tables-only file, rather than
2343 hard-wiring them into your application. The jpeg_read_header() call is
2344 sufficient to read a tables-only file. You must pass a second parameter of
2345 FALSE to indicate that you do not require an image to be present. Thus, the
2348 create JPEG decompression object
2349 set source to tables-only file
2350 jpeg_read_header(&cinfo, FALSE);
2351 set source to abbreviated image file
2352 jpeg_read_header(&cinfo, TRUE);
2353 set decompression parameters
2354 jpeg_start_decompress(&cinfo);
2356 jpeg_finish_decompress(&cinfo);
2358 In some cases, you may want to read a file without knowing whether it contains
2359 an image or just tables. In that case, pass FALSE and check the return value
2360 from jpeg_read_header(): it will be JPEG_HEADER_OK if an image was found,
2361 JPEG_HEADER_TABLES_ONLY if only tables were found. (A third return value,
2362 JPEG_SUSPENDED, is possible when using a suspending data source manager.)
2363 Note that jpeg_read_header() will not complain if you read an abbreviated
2364 image for which you haven't loaded the missing tables; the missing-table check
2365 occurs later, in jpeg_start_decompress().
2368 It is possible to read a series of images from a single source file by
2369 repeating the jpeg_read_header() ... jpeg_finish_decompress() sequence,
2370 without releasing/recreating the JPEG object or the data source module.
2371 (If you did reinitialize, any partial bufferload left in the data source
2372 buffer at the end of one image would be discarded, causing you to lose the
2373 start of the next image.) When you use this method, stored tables are
2374 automatically carried forward, so some of the images can be abbreviated images
2375 that depend on tables from earlier images.
2377 If you intend to write a series of images into a single destination file,
2378 you might want to make a specialized data destination module that doesn't
2379 flush the output buffer at term_destination() time. This would speed things
2380 up by some trifling amount. Of course, you'd need to remember to flush the
2381 buffer after the last image. You can make the later images be abbreviated
2382 ones by passing FALSE to jpeg_start_compress().
2388 Some applications may need to insert or extract special data in the JPEG
2389 datastream. The JPEG standard provides marker types "COM" (comment) and
2390 "APP0" through "APP15" (application) to hold application-specific data.
2391 Unfortunately, the use of these markers is not specified by the standard.
2392 COM markers are fairly widely used to hold user-supplied text. The JFIF file
2393 format spec uses APP0 markers with specified initial strings to hold certain
2394 data. Adobe applications use APP14 markers beginning with the string "Adobe"
2395 for miscellaneous data. Other APPn markers are rarely seen, but might
2396 contain almost anything.
2398 If you wish to store user-supplied text, we recommend you use COM markers
2399 and place readable 7-bit ASCII text in them. Newline conventions are not
2400 standardized --- expect to find LF (Unix style), CR/LF (DOS style), or CR
2401 (Mac style). A robust COM reader should be able to cope with random binary
2402 garbage, including nulls, since some applications generate COM markers
2403 containing non-ASCII junk. (But yours should not be one of them.)
2405 For program-supplied data, use an APPn marker, and be sure to begin it with an
2406 identifying string so that you can tell whether the marker is actually yours.
2407 It's probably best to avoid using APP0 or APP14 for any private markers.
2408 (NOTE: the upcoming SPIFF standard will use APP8 markers; we recommend you
2409 not use APP8 markers for any private purposes, either.)
2411 Keep in mind that at most 65533 bytes can be put into one marker, but you
2412 can have as many markers as you like.
2414 By default, the IJG compression library will write a JFIF APP0 marker if the
2415 selected JPEG colorspace is grayscale or YCbCr, or an Adobe APP14 marker if
2416 the selected colorspace is RGB, CMYK, or YCCK. You can disable this, but
2417 we don't recommend it. The decompression library will recognize JFIF and
2418 Adobe markers and will set the JPEG colorspace properly when one is found.
2421 You can write special markers immediately following the datastream header by
2422 calling jpeg_write_marker() after jpeg_start_compress() and before the first
2423 call to jpeg_write_scanlines(). When you do this, the markers appear after
2424 the SOI and the JFIF APP0 and Adobe APP14 markers (if written), but before
2425 all else. Specify the marker type parameter as "JPEG_COM" for COM or
2426 "JPEG_APP0 + n" for APPn. (Actually, jpeg_write_marker will let you write
2427 any marker type, but we don't recommend writing any other kinds of marker.)
2428 For example, to write a user comment string pointed to by comment_text:
2429 jpeg_write_marker(cinfo, JPEG_COM, comment_text, strlen(comment_text));
2431 If it's not convenient to store all the marker data in memory at once,
2432 you can instead call jpeg_write_m_header() followed by multiple calls to
2433 jpeg_write_m_byte(). If you do it this way, it's your responsibility to
2434 call jpeg_write_m_byte() exactly the number of times given in the length
2435 parameter to jpeg_write_m_header(). (This method lets you empty the
2436 output buffer partway through a marker, which might be important when
2437 using a suspending data destination module. In any case, if you are using
2438 a suspending destination, you should flush its buffer after inserting
2439 any special markers. See "I/O suspension".)
2441 Or, if you prefer to synthesize the marker byte sequence yourself,
2442 you can just cram it straight into the data destination module.
2444 If you are writing JFIF 1.02 extension markers (thumbnail images), don't
2445 forget to set cinfo.JFIF_minor_version = 2 so that the encoder will write the
2446 correct JFIF version number in the JFIF header marker. The library's default
2447 is to write version 1.01, but that's wrong if you insert any 1.02 extension
2448 markers. (We could probably get away with just defaulting to 1.02, but there
2449 used to be broken decoders that would complain about unknown minor version
2450 numbers. To reduce compatibility risks it's safest not to write 1.02 unless
2451 you are actually using 1.02 extensions.)
2454 When reading, two methods of handling special markers are available:
2455 1. You can ask the library to save the contents of COM and/or APPn markers
2456 into memory, and then examine them at your leisure afterwards.
2457 2. You can supply your own routine to process COM and/or APPn markers
2458 on-the-fly as they are read.
2459 The first method is simpler to use, especially if you are using a suspending
2460 data source; writing a marker processor that copes with input suspension is
2461 not easy (consider what happens if the marker is longer than your available
2462 input buffer). However, the second method conserves memory since the marker
2463 data need not be kept around after it's been processed.
2465 For either method, you'd normally set up marker handling after creating a
2466 decompression object and before calling jpeg_read_header(), because the
2467 markers of interest will typically be near the head of the file and so will
2468 be scanned by jpeg_read_header. Once you've established a marker handling
2469 method, it will be used for the life of that decompression object
2470 (potentially many datastreams), unless you change it. Marker handling is
2471 determined separately for COM markers and for each APPn marker code.
2474 To save the contents of special markers in memory, call
2475 jpeg_save_markers(cinfo, marker_code, length_limit)
2476 where marker_code is the marker type to save, JPEG_COM or JPEG_APP0+n.
2477 (To arrange to save all the special marker types, you need to call this
2478 routine 17 times, for COM and APP0-APP15.) If the incoming marker is longer
2479 than length_limit data bytes, only length_limit bytes will be saved; this
2480 parameter allows you to avoid chewing up memory when you only need to see the
2481 first few bytes of a potentially large marker. If you want to save all the
2482 data, set length_limit to 0xFFFF; that is enough since marker lengths are only
2483 16 bits. As a special case, setting length_limit to 0 prevents that marker
2484 type from being saved at all. (That is the default behavior, in fact.)
2486 After jpeg_read_header() completes, you can examine the special markers by
2487 following the cinfo->marker_list pointer chain. All the special markers in
2488 the file appear in this list, in order of their occurrence in the file (but
2489 omitting any markers of types you didn't ask for). Both the original data
2490 length and the saved data length are recorded for each list entry; the latter
2491 will not exceed length_limit for the particular marker type. Note that these
2492 lengths exclude the marker length word, whereas the stored representation
2493 within the JPEG file includes it. (Hence the maximum data length is really
2496 It is possible that additional special markers appear in the file beyond the
2497 SOS marker at which jpeg_read_header stops; if so, the marker list will be
2498 extended during reading of the rest of the file. This is not expected to be
2499 common, however. If you are short on memory you may want to reset the length
2500 limit to zero for all marker types after finishing jpeg_read_header, to
2501 ensure that the max_memory_to_use setting cannot be exceeded due to addition
2504 The marker list remains stored until you call jpeg_finish_decompress or
2505 jpeg_abort, at which point the memory is freed and the list is set to empty.
2506 (jpeg_destroy also releases the storage, of course.)
2508 Note that the library is internally interested in APP0 and APP14 markers;
2509 if you try to set a small nonzero length limit on these types, the library
2510 will silently force the length up to the minimum it wants. (But you can set
2511 a zero length limit to prevent them from being saved at all.) Also, in a
2512 16-bit environment, the maximum length limit may be constrained to less than
2513 65533 by malloc() limitations. It is therefore best not to assume that the
2514 effective length limit is exactly what you set it to be.
2517 If you want to supply your own marker-reading routine, you do it by calling
2518 jpeg_set_marker_processor(). A marker processor routine must have the
2520 boolean jpeg_marker_parser_method (j_decompress_ptr cinfo)
2521 Although the marker code is not explicitly passed, the routine can find it
2522 in cinfo->unread_marker. At the time of call, the marker proper has been
2523 read from the data source module. The processor routine is responsible for
2524 reading the marker length word and the remaining parameter bytes, if any.
2525 Return TRUE to indicate success. (FALSE should be returned only if you are
2526 using a suspending data source and it tells you to suspend. See the standard
2527 marker processors in jdmarker.c for appropriate coding methods if you need to
2528 use a suspending data source.)
2530 If you override the default APP0 or APP14 processors, it is up to you to
2531 recognize JFIF and Adobe markers if you want colorspace recognition to occur
2532 properly. We recommend copying and extending the default processors if you
2533 want to do that. (A better idea is to save these marker types for later
2534 examination by calling jpeg_save_markers(); that method doesn't interfere
2535 with the library's own processing of these markers.)
2537 jpeg_set_marker_processor() and jpeg_save_markers() are mutually exclusive
2538 --- if you call one it overrides any previous call to the other, for the
2539 particular marker type specified.
2541 A simple example of an external COM processor can be found in djpeg.c.
2542 Also, see jpegtran.c for an example of using jpeg_save_markers.
2545 Raw (downsampled) image data
2546 ----------------------------
2548 Some applications need to supply already-downsampled image data to the JPEG
2549 compressor, or to receive raw downsampled data from the decompressor. The
2550 library supports this requirement by allowing the application to write or
2551 read raw data, bypassing the normal preprocessing or postprocessing steps.
2552 The interface is different from the standard one and is somewhat harder to
2553 use. If your interest is merely in bypassing color conversion, we recommend
2554 that you use the standard interface and simply set jpeg_color_space =
2555 in_color_space (or jpeg_color_space = out_color_space for decompression).
2556 The mechanism described in this section is necessary only to supply or
2557 receive downsampled image data, in which not all components have the same
2561 To compress raw data, you must supply the data in the colorspace to be used
2562 in the JPEG file (please read the earlier section on Special color spaces)
2563 and downsampled to the sampling factors specified in the JPEG parameters.
2564 You must supply the data in the format used internally by the JPEG library,
2565 namely a JSAMPIMAGE array. This is an array of pointers to two-dimensional
2566 arrays, each of type JSAMPARRAY. Each 2-D array holds the values for one
2567 color component. This structure is necessary since the components are of
2568 different sizes. If the image dimensions are not a multiple of the MCU size,
2569 you must also pad the data correctly (usually, this is done by replicating
2570 the last column and/or row). The data must be padded to a multiple of a DCT
2571 block in each component: that is, each downsampled row must contain a
2572 multiple of 8 valid samples, and there must be a multiple of 8 sample rows
2573 for each component. (For applications such as conversion of digital TV
2574 images, the standard image size is usually a multiple of the DCT block size,
2575 so that no padding need actually be done.)
2577 The procedure for compression of raw data is basically the same as normal
2578 compression, except that you call jpeg_write_raw_data() in place of
2579 jpeg_write_scanlines(). Before calling jpeg_start_compress(), you must do
2581 * Set cinfo->raw_data_in to TRUE. (It is set FALSE by jpeg_set_defaults().)
2582 This notifies the library that you will be supplying raw data.
2583 * Ensure jpeg_color_space is correct --- an explicit jpeg_set_colorspace()
2584 call is a good idea. Note that since color conversion is bypassed,
2585 in_color_space is ignored, except that jpeg_set_defaults() uses it to
2586 choose the default jpeg_color_space setting.
2587 * Ensure the sampling factors, cinfo->comp_info[i].h_samp_factor and
2588 cinfo->comp_info[i].v_samp_factor, are correct. Since these indicate the
2589 dimensions of the data you are supplying, it's wise to set them
2590 explicitly, rather than assuming the library's defaults are what you want.
2592 To pass raw data to the library, call jpeg_write_raw_data() in place of
2593 jpeg_write_scanlines(). The two routines work similarly except that
2594 jpeg_write_raw_data takes a JSAMPIMAGE data array rather than JSAMPARRAY.
2595 The scanlines count passed to and returned from jpeg_write_raw_data is
2596 measured in terms of the component with the largest v_samp_factor.
2598 jpeg_write_raw_data() processes one MCU row per call, which is to say
2599 v_samp_factor*DCTSIZE sample rows of each component. The passed num_lines
2600 value must be at least max_v_samp_factor*DCTSIZE, and the return value will
2601 be exactly that amount (or possibly some multiple of that amount, in future
2602 library versions). This is true even on the last call at the bottom of the
2603 image; don't forget to pad your data as necessary.
2605 The required dimensions of the supplied data can be computed for each
2607 cinfo->comp_info[i].width_in_blocks*DCTSIZE samples per row
2608 cinfo->comp_info[i].height_in_blocks*DCTSIZE rows in image
2609 after jpeg_start_compress() has initialized those fields. If the valid data
2610 is smaller than this, it must be padded appropriately. For some sampling
2611 factors and image sizes, additional dummy DCT blocks are inserted to make
2612 the image a multiple of the MCU dimensions. The library creates such dummy
2613 blocks itself; it does not read them from your supplied data. Therefore you
2614 need never pad by more than DCTSIZE samples. An example may help here.
2615 Assume 2h2v downsampling of YCbCr data, that is
2616 cinfo->comp_info[0].h_samp_factor = 2 for Y
2617 cinfo->comp_info[0].v_samp_factor = 2
2618 cinfo->comp_info[1].h_samp_factor = 1 for Cb
2619 cinfo->comp_info[1].v_samp_factor = 1
2620 cinfo->comp_info[2].h_samp_factor = 1 for Cr
2621 cinfo->comp_info[2].v_samp_factor = 1
2622 and suppose that the nominal image dimensions (cinfo->image_width and
2623 cinfo->image_height) are 101x101 pixels. Then jpeg_start_compress() will
2624 compute downsampled_width = 101 and width_in_blocks = 13 for Y,
2625 downsampled_width = 51 and width_in_blocks = 7 for Cb and Cr (and the same
2626 for the height fields). You must pad the Y data to at least 13*8 = 104
2627 columns and rows, the Cb/Cr data to at least 7*8 = 56 columns and rows. The
2628 MCU height is max_v_samp_factor = 2 DCT rows so you must pass at least 16
2629 scanlines on each call to jpeg_write_raw_data(), which is to say 16 actual
2630 sample rows of Y and 8 each of Cb and Cr. A total of 7 MCU rows are needed,
2631 so you must pass a total of 7*16 = 112 "scanlines". The last DCT block row
2632 of Y data is dummy, so it doesn't matter what you pass for it in the data
2633 arrays, but the scanlines count must total up to 112 so that all of the Cb
2634 and Cr data gets passed.
2636 Output suspension is supported with raw-data compression: if the data
2637 destination module suspends, jpeg_write_raw_data() will return 0.
2638 In this case the same data rows must be passed again on the next call.
2641 Decompression with raw data output implies bypassing all postprocessing:
2642 you cannot ask for rescaling or color quantization, for instance. More
2643 seriously, you must deal with the color space and sampling factors present in
2644 the incoming file. If your application only handles, say, 2h1v YCbCr data,
2645 you must check for and fail on other color spaces or other sampling factors.
2646 The library will not convert to a different color space for you.
2648 To obtain raw data output, set cinfo->raw_data_out = TRUE before
2649 jpeg_start_decompress() (it is set FALSE by jpeg_read_header()). Be sure to
2650 verify that the color space and sampling factors are ones you can handle.
2651 Then call jpeg_read_raw_data() in place of jpeg_read_scanlines(). The
2652 decompression process is otherwise the same as usual.
2654 jpeg_read_raw_data() returns one MCU row per call, and thus you must pass a
2655 buffer of at least max_v_samp_factor*DCTSIZE scanlines (scanline counting is
2656 the same as for raw-data compression). The buffer you pass must be large
2657 enough to hold the actual data plus padding to DCT-block boundaries. As with
2658 compression, any entirely dummy DCT blocks are not processed so you need not
2659 allocate space for them, but the total scanline count includes them. The
2660 above example of computing buffer dimensions for raw-data compression is
2661 equally valid for decompression.
2663 Input suspension is supported with raw-data decompression: if the data source
2664 module suspends, jpeg_read_raw_data() will return 0. You can also use
2665 buffered-image mode to read raw data in multiple passes.
2668 Really raw data: DCT coefficients
2669 ---------------------------------
2671 It is possible to read or write the contents of a JPEG file as raw DCT
2672 coefficients. This facility is mainly intended for use in lossless
2673 transcoding between different JPEG file formats. Other possible applications
2674 include lossless cropping of a JPEG image, lossless reassembly of a
2675 multi-strip or multi-tile TIFF/JPEG file into a single JPEG datastream, etc.
2677 To read the contents of a JPEG file as DCT coefficients, open the file and do
2678 jpeg_read_header() as usual. But instead of calling jpeg_start_decompress()
2679 and jpeg_read_scanlines(), call jpeg_read_coefficients(). This will read the
2680 entire image into a set of virtual coefficient-block arrays, one array per
2681 component. The return value is a pointer to an array of virtual-array
2682 descriptors. Each virtual array can be accessed directly using the JPEG
2683 memory manager's access_virt_barray method (see Memory management, below,
2684 and also read structure.txt's discussion of virtual array handling). Or,
2685 for simple transcoding to a different JPEG file format, the array list can
2686 just be handed directly to jpeg_write_coefficients().
2688 Each block in the block arrays contains quantized coefficient values in
2689 normal array order (not JPEG zigzag order). The block arrays contain only
2690 DCT blocks containing real data; any entirely-dummy blocks added to fill out
2691 interleaved MCUs at the right or bottom edges of the image are discarded
2692 during reading and are not stored in the block arrays. (The size of each
2693 block array can be determined from the width_in_blocks and height_in_blocks
2694 fields of the component's comp_info entry.) This is also the data format
2695 expected by jpeg_write_coefficients().
2697 When you are done using the virtual arrays, call jpeg_finish_decompress()
2698 to release the array storage and return the decompression object to an idle
2699 state; or just call jpeg_destroy() if you don't need to reuse the object.
2701 If you use a suspending data source, jpeg_read_coefficients() will return
2702 NULL if it is forced to suspend; a non-NULL return value indicates successful
2703 completion. You need not test for a NULL return value when using a
2704 non-suspending data source.
2706 It is also possible to call jpeg_read_coefficients() to obtain access to the
2707 decoder's coefficient arrays during a normal decode cycle in buffered-image
2708 mode. This frammish might be useful for progressively displaying an incoming
2709 image and then re-encoding it without loss. To do this, decode in buffered-
2710 image mode as discussed previously, then call jpeg_read_coefficients() after
2711 the last jpeg_finish_output() call. The arrays will be available for your use
2712 until you call jpeg_finish_decompress().
2715 To write the contents of a JPEG file as DCT coefficients, you must provide
2716 the DCT coefficients stored in virtual block arrays. You can either pass
2717 block arrays read from an input JPEG file by jpeg_read_coefficients(), or
2718 allocate virtual arrays from the JPEG compression object and fill them
2719 yourself. In either case, jpeg_write_coefficients() is substituted for
2720 jpeg_start_compress() and jpeg_write_scanlines(). Thus the sequence is
2721 * Create compression object
2722 * Set all compression parameters as necessary
2723 * Request virtual arrays if needed
2724 * jpeg_write_coefficients()
2725 * jpeg_finish_compress()
2726 * Destroy or re-use compression object
2727 jpeg_write_coefficients() is passed a pointer to an array of virtual block
2728 array descriptors; the number of arrays is equal to cinfo.num_components.
2730 The virtual arrays need only have been requested, not realized, before
2731 jpeg_write_coefficients() is called. A side-effect of
2732 jpeg_write_coefficients() is to realize any virtual arrays that have been
2733 requested from the compression object's memory manager. Thus, when obtaining
2734 the virtual arrays from the compression object, you should fill the arrays
2735 after calling jpeg_write_coefficients(). The data is actually written out
2736 when you call jpeg_finish_compress(); jpeg_write_coefficients() only writes
2739 When writing raw DCT coefficients, it is crucial that the JPEG quantization
2740 tables and sampling factors match the way the data was encoded, or the
2741 resulting file will be invalid. For transcoding from an existing JPEG file,
2742 we recommend using jpeg_copy_critical_parameters(). This routine initializes
2743 all the compression parameters to default values (like jpeg_set_defaults()),
2744 then copies the critical information from a source decompression object.
2745 The decompression object should have just been used to read the entire
2746 JPEG input file --- that is, it should be awaiting jpeg_finish_decompress().
2748 jpeg_write_coefficients() marks all tables stored in the compression object
2749 as needing to be written to the output file (thus, it acts like
2750 jpeg_start_compress(cinfo, TRUE)). This is for safety's sake, to avoid
2751 emitting abbreviated JPEG files by accident. If you really want to emit an
2752 abbreviated JPEG file, call jpeg_suppress_tables(), or set the tables'
2753 individual sent_table flags, between calling jpeg_write_coefficients() and
2754 jpeg_finish_compress().
2760 Some applications may need to regain control from the JPEG library every so
2761 often. The typical use of this feature is to produce a percent-done bar or
2762 other progress display. (For a simple example, see cjpeg.c or djpeg.c.)
2763 Although you do get control back frequently during the data-transferring pass
2764 (the jpeg_read_scanlines or jpeg_write_scanlines loop), any additional passes
2765 will occur inside jpeg_finish_compress or jpeg_start_decompress; those
2766 routines may take a long time to execute, and you don't get control back
2767 until they are done.
2769 You can define a progress-monitor routine which will be called periodically
2770 by the library. No guarantees are made about how often this call will occur,
2771 so we don't recommend you use it for mouse tracking or anything like that.
2772 At present, a call will occur once per MCU row, scanline, or sample row
2773 group, whichever unit is convenient for the current processing mode; so the
2774 wider the image, the longer the time between calls. During the data
2775 transferring pass, only one call occurs per call of jpeg_read_scanlines or
2776 jpeg_write_scanlines, so don't pass a large number of scanlines at once if
2777 you want fine resolution in the progress count. (If you really need to use
2778 the callback mechanism for time-critical tasks like mouse tracking, you could
2779 insert additional calls inside some of the library's inner loops.)
2781 To establish a progress-monitor callback, create a struct jpeg_progress_mgr,
2782 fill in its progress_monitor field with a pointer to your callback routine,
2783 and set cinfo->progress to point to the struct. The callback will be called
2784 whenever cinfo->progress is non-NULL. (This pointer is set to NULL by
2785 jpeg_create_compress or jpeg_create_decompress; the library will not change
2786 it thereafter. So if you allocate dynamic storage for the progress struct,
2787 make sure it will live as long as the JPEG object does. Allocating from the
2788 JPEG memory manager with lifetime JPOOL_PERMANENT will work nicely.) You
2789 can use the same callback routine for both compression and decompression.
2791 The jpeg_progress_mgr struct contains four fields which are set by the library:
2792 long pass_counter; /* work units completed in this pass */
2793 long pass_limit; /* total number of work units in this pass */
2794 int completed_passes; /* passes completed so far */
2795 int total_passes; /* total number of passes expected */
2796 During any one pass, pass_counter increases from 0 up to (not including)
2797 pass_limit; the step size is usually but not necessarily 1. The pass_limit
2798 value may change from one pass to another. The expected total number of
2799 passes is in total_passes, and the number of passes already completed is in
2800 completed_passes. Thus the fraction of work completed may be estimated as
2801 completed_passes + (pass_counter/pass_limit)
2802 --------------------------------------------
2804 ignoring the fact that the passes may not be equal amounts of work.
2806 When decompressing, pass_limit can even change within a pass, because it
2807 depends on the number of scans in the JPEG file, which isn't always known in
2808 advance. The computed fraction-of-work-done may jump suddenly (if the library
2809 discovers it has overestimated the number of scans) or even decrease (in the
2810 opposite case). It is not wise to put great faith in the work estimate.
2812 When using the decompressor's buffered-image mode, the progress monitor work
2813 estimate is likely to be completely unhelpful, because the library has no way
2814 to know how many output passes will be demanded of it. Currently, the library
2815 sets total_passes based on the assumption that there will be one more output
2816 pass if the input file end hasn't yet been read (jpeg_input_complete() isn't
2817 TRUE), but no more output passes if the file end has been reached when the
2818 output pass is started. This means that total_passes will rise as additional
2819 output passes are requested. If you have a way of determining the input file
2820 size, estimating progress based on the fraction of the file that's been read
2821 will probably be more useful than using the library's value.
2827 This section covers some key facts about the JPEG library's built-in memory
2828 manager. For more info, please read structure.txt's section about the memory
2829 manager, and consult the source code if necessary.
2831 All memory and temporary file allocation within the library is done via the
2832 memory manager. If necessary, you can replace the "back end" of the memory
2833 manager to control allocation yourself (for example, if you don't want the
2834 library to use malloc() and free() for some reason).
2836 Some data is allocated "permanently" and will not be freed until the JPEG
2837 object is destroyed. Most data is allocated "per image" and is freed by
2838 jpeg_finish_compress, jpeg_finish_decompress, or jpeg_abort. You can call the
2839 memory manager yourself to allocate structures that will automatically be
2840 freed at these times. Typical code for this is
2841 ptr = (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE, size);
2842 Use JPOOL_PERMANENT to get storage that lasts as long as the JPEG object.
2843 Use alloc_large instead of alloc_small for anything bigger than a few Kbytes.
2844 There are also alloc_sarray and alloc_barray routines that automatically
2845 build 2-D sample or block arrays.
2847 The library's minimum space requirements to process an image depend on the
2848 image's width, but not on its height, because the library ordinarily works
2849 with "strip" buffers that are as wide as the image but just a few rows high.
2850 Some operating modes (eg, two-pass color quantization) require full-image
2851 buffers. Such buffers are treated as "virtual arrays": only the current strip
2852 need be in memory, and the rest can be swapped out to a temporary file.
2854 If you use the simplest memory manager back end (jmemnobs.c), then no
2855 temporary files are used; virtual arrays are simply malloc()'d. Images bigger
2856 than memory can be processed only if your system supports virtual memory.
2857 The other memory manager back ends support temporary files of various flavors
2858 and thus work in machines without virtual memory. They may also be useful on
2859 Unix machines if you need to process images that exceed available swap space.
2861 When using temporary files, the library will make the in-memory buffers for
2862 its virtual arrays just big enough to stay within a "maximum memory" setting.
2863 Your application can set this limit by setting cinfo->mem->max_memory_to_use
2864 after creating the JPEG object. (Of course, there is still a minimum size for
2865 the buffers, so the max-memory setting is effective only if it is bigger than
2866 the minimum space needed.) If you allocate any large structures yourself, you
2867 must allocate them before jpeg_start_compress() or jpeg_start_decompress() in
2868 order to have them counted against the max memory limit. Also keep in mind
2869 that space allocated with alloc_small() is ignored, on the assumption that
2870 it's too small to be worth worrying about; so a reasonable safety margin
2871 should be left when setting max_memory_to_use.
2877 Working memory requirements while performing compression or decompression
2878 depend on image dimensions, image characteristics (such as colorspace and
2879 JPEG process), and operating mode (application-selected options).
2881 As of v6b, the decompressor requires:
2882 1. About 24K in more-or-less-fixed-size data. This varies a bit depending
2883 on operating mode and image characteristics (particularly color vs.
2884 grayscale), but it doesn't depend on image dimensions.
2885 2. Strip buffers (of size proportional to the image width) for IDCT and
2886 upsampling results. The worst case for commonly used sampling factors
2887 is about 34 bytes * width in pixels for a color image. A grayscale image
2888 only needs about 8 bytes per pixel column.
2889 3. A full-image DCT coefficient buffer is needed to decode a multi-scan JPEG
2890 file (including progressive JPEGs), or whenever you select buffered-image
2891 mode. This takes 2 bytes/coefficient. At typical 2x2 sampling, that's
2892 3 bytes per pixel for a color image. Worst case (1x1 sampling) requires
2893 6 bytes/pixel. For grayscale, figure 2 bytes/pixel.
2894 4. To perform 2-pass color quantization, the decompressor also needs a
2895 128K color lookup table and a full-image pixel buffer (3 bytes/pixel).
2896 This does not count any memory allocated by the application, such as a
2897 buffer to hold the final output image.
2899 The above figures are valid for 8-bit JPEG data precision and a machine with
2900 32-bit ints. For 12-bit JPEG data, double the size of the strip buffers and
2901 quantization pixel buffer. The "fixed-size" data will be somewhat smaller
2902 with 16-bit ints, larger with 64-bit ints. Also, CMYK or other unusual
2903 color spaces will require different amounts of space.
2905 The full-image coefficient and pixel buffers, if needed at all, do not
2906 have to be fully RAM resident; you can have the library use temporary
2907 files instead when the total memory usage would exceed a limit you set.
2908 (But if your OS supports virtual memory, it's probably better to just use
2909 jmemnobs and let the OS do the swapping.)
2911 The compressor's memory requirements are similar, except that it has no need
2912 for color quantization. Also, it needs a full-image DCT coefficient buffer
2913 if Huffman-table optimization is asked for, even if progressive mode is not
2916 If you need more detailed information about memory usage in a particular
2917 situation, you can enable the MEM_STATS code in jmemmgr.c.
2920 Library compile-time options
2921 ----------------------------
2923 A number of compile-time options are available by modifying jmorecfg.h.
2925 The JPEG standard provides for both the baseline 8-bit DCT process and
2926 a 12-bit DCT process. The IJG code supports 12-bit lossy JPEG if you define
2927 BITS_IN_JSAMPLE as 12 rather than 8. Note that this causes JSAMPLE to be
2928 larger than a char, so it affects the surrounding application's image data.
2929 The sample applications cjpeg and djpeg can support 12-bit mode only for PPM
2930 and GIF file formats; you must disable the other file formats to compile a
2931 12-bit cjpeg or djpeg. (install.txt has more information about that.)
2932 At present, a 12-bit library can handle *only* 12-bit images, not both
2935 Note that a 12-bit library always compresses in Huffman optimization mode,
2936 in order to generate valid Huffman tables. This is necessary because our
2937 default Huffman tables only cover 8-bit data. If you need to output 12-bit
2938 files in one pass, you'll have to supply suitable default Huffman tables.
2939 You may also want to supply your own DCT quantization tables; the existing
2940 quality-scaling code has been developed for 8-bit use, and probably doesn't
2941 generate especially good tables for 12-bit.
2943 The maximum number of components (color channels) in the image is determined
2944 by MAX_COMPONENTS. The JPEG standard allows up to 255 components, but we
2945 expect that few applications will need more than four or so.
2947 On machines with unusual data type sizes, you may be able to improve
2948 performance or reduce memory space by tweaking the various typedefs in
2949 jmorecfg.h. In particular, on some RISC CPUs, access to arrays of "short"s
2950 is quite slow; consider trading memory for speed by making JCOEF, INT16, and
2951 UINT16 be "int" or "unsigned int". UINT8 is also a candidate to become int.
2952 You probably don't want to make JSAMPLE be int unless you have lots of memory
2955 You can reduce the size of the library by compiling out various optional
2956 functions. To do this, undefine xxx_SUPPORTED symbols as necessary.
2958 You can also save a few K by not having text error messages in the library;
2959 the standard error message table occupies about 5Kb. This is particularly
2960 reasonable for embedded applications where there's no good way to display
2961 a message anyway. To do this, remove the creation of the message table
2962 (jpeg_std_message_table[]) from jerror.c, and alter format_message to do
2963 something reasonable without it. You could output the numeric value of the
2964 message code number, for example. If you do this, you can also save a couple
2965 more K by modifying the TRACEMSn() macros in jerror.h to expand to nothing;
2966 you don't need trace capability anyway, right?
2969 Portability considerations
2970 --------------------------
2972 The JPEG library has been written to be extremely portable; the sample
2973 applications cjpeg and djpeg are slightly less so. This section summarizes
2974 the design goals in this area. (If you encounter any bugs that cause the
2975 library to be less portable than is claimed here, we'd appreciate hearing
2978 The code works fine on ANSI C and C++ compilers, using any of the popular
2979 system include file setups, and some not-so-popular ones too.
2981 The code is not dependent on the exact sizes of the C data types. As
2982 distributed, we make the assumptions that
2983 char is at least 8 bits wide
2984 short is at least 16 bits wide
2985 int is at least 16 bits wide
2986 long is at least 32 bits wide
2987 (These are the minimum requirements of the ANSI C standard.) Wider types will
2988 work fine, although memory may be used inefficiently if char is much larger
2989 than 8 bits or short is much bigger than 16 bits. The code should work
2990 equally well with 16- or 32-bit ints.
2992 In a system where these assumptions are not met, you may be able to make the
2993 code work by modifying the typedefs in jmorecfg.h. However, you will probably
2994 have difficulty if int is less than 16 bits wide, since references to plain
2995 int abound in the code.
2997 char can be either signed or unsigned, although the code runs faster if an
2998 unsigned char type is available. If char is wider than 8 bits, you will need
2999 to redefine JOCTET and/or provide custom data source/destination managers so
3000 that JOCTET represents exactly 8 bits of data on external storage.
3002 The JPEG library proper does not assume ASCII representation of characters.
3003 But some of the image file I/O modules in cjpeg/djpeg do have ASCII
3004 dependencies in file-header manipulation; so does cjpeg's select_file_type()
3007 The JPEG library does not rely heavily on the C library. In particular, C
3008 stdio is used only by the data source/destination modules and the error
3009 handler, all of which are application-replaceable. (cjpeg/djpeg are more
3010 heavily dependent on stdio.) malloc and free are called only from the memory
3011 manager "back end" module, so you can use a different memory allocator by
3012 replacing that one file.
3014 More info about porting the code may be gleaned by reading jconfig.txt,
3015 jmorecfg.h, and jinclude.h.