4 libjpeg-turbo is a JPEG image codec that uses SIMD instructions to accelerate
5 baseline JPEG compression and decompression on x86, x86-64, Arm, PowerPC, and
6 MIPS systems, as well as progressive JPEG compression on x86, x86-64, and Arm
7 systems. On such systems, libjpeg-turbo is generally 2-6x as fast as libjpeg,
8 all else being equal. On other types of systems, libjpeg-turbo can still
9 outperform libjpeg by a significant amount, by virtue of its highly-optimized
10 Huffman coding routines. In many cases, the performance of libjpeg-turbo
11 rivals that of proprietary high-speed JPEG codecs.
13 libjpeg-turbo implements both the traditional libjpeg API as well as the less
14 powerful but more straightforward TurboJPEG API. libjpeg-turbo also features
15 colorspace extensions that allow it to compress from/decompress to 32-bit and
16 big-endian pixel buffers (RGBX, XBGR, etc.), as well as a full-featured Java
19 libjpeg-turbo was originally based on libjpeg/SIMD, an MMX-accelerated
20 derivative of libjpeg v6b developed by Miyasaka Masaru. The TigerVNC and
21 VirtualGL projects made numerous enhancements to the codec in 2009, and in
22 early 2010, libjpeg-turbo spun off into an independent project, with the goal
23 of making high-speed JPEG compression/decompression technology available to a
24 broader range of users and developers. libjpeg-turbo is an ISO/IEC and ITU-T
25 reference implementation of the JPEG standard.
27 More information about libjpeg-turbo can be found at
28 <https://libjpeg-turbo.org>.
34 libjpeg-turbo is an independent open source project, but we rely on patronage
35 and funded development in order to maintain that independence. The easiest way
36 to ensure that libjpeg-turbo remains community-focused and free of any one
37 organization's agenda is to
38 [sponsor our project through GitHub](https://github.com/sponsors/libjpeg-turbo).
39 All sponsorship money goes directly toward funding the labor necessary to
40 maintain libjpeg-turbo, support the user community, and implement bug fixes and
41 strategically important features.
43 [![Sponsor libjpeg-turbo](https://img.shields.io/github/sponsors/libjpeg-turbo?label=Sponsor&logo=GitHub)](https://github.com/sponsors/libjpeg-turbo)
49 libjpeg-turbo is covered by three compatible BSD-style open source licenses.
50 Refer to [LICENSE.md](LICENSE.md) for a roll-up of license terms.
53 Building libjpeg-turbo
54 ======================
56 Refer to [BUILDING.md](BUILDING.md) for complete instructions.
62 libjpeg-turbo includes two APIs that can be used to compress and decompress
65 - **TurboJPEG API**<br>
66 This API provides an easy-to-use interface for compressing and decompressing
67 JPEG images in memory. It also provides some functionality that would not be
68 straightforward to achieve using the underlying libjpeg API, such as
69 generating planar YUV images and performing multiple simultaneous lossless
70 transforms on an image. The Java interface for libjpeg-turbo is written on
71 top of the TurboJPEG API. The TurboJPEG API is recommended for first-time
72 users of libjpeg-turbo. Refer to [tjexample.c](tjexample.c) and
73 [TJExample.java](java/TJExample.java) for examples of its usage and to
74 <http://libjpeg-turbo.org/Documentation/Documentation> for API documentation.
77 This is the de facto industry-standard API for compressing and decompressing
78 JPEG images. It is more difficult to use than the TurboJPEG API but also
79 more powerful. The libjpeg API implementation in libjpeg-turbo is both
80 API/ABI-compatible and mathematically compatible with libjpeg v6b. It can
81 also optionally be configured to be API/ABI-compatible with libjpeg v7 and v8
82 (see below.) Refer to [cjpeg.c](cjpeg.c) and [djpeg.c](djpeg.c) for examples
83 of its usage and to [libjpeg.txt](libjpeg.txt) for API documentation.
85 There is no significant performance advantage to either API when both are used
86 to perform similar operations.
91 libjpeg-turbo includes extensions that allow JPEG images to be compressed
92 directly from (and decompressed directly to) buffers that use BGR, BGRX,
93 RGBX, XBGR, and XRGB pixel ordering. This is implemented with ten new
96 JCS_EXT_RGB /* red/green/blue */
97 JCS_EXT_RGBX /* red/green/blue/x */
98 JCS_EXT_BGR /* blue/green/red */
99 JCS_EXT_BGRX /* blue/green/red/x */
100 JCS_EXT_XBGR /* x/blue/green/red */
101 JCS_EXT_XRGB /* x/red/green/blue */
102 JCS_EXT_RGBA /* red/green/blue/alpha */
103 JCS_EXT_BGRA /* blue/green/red/alpha */
104 JCS_EXT_ABGR /* alpha/blue/green/red */
105 JCS_EXT_ARGB /* alpha/red/green/blue */
107 Setting `cinfo.in_color_space` (compression) or `cinfo.out_color_space`
108 (decompression) to one of these values will cause libjpeg-turbo to read the
109 red, green, and blue values from (or write them to) the appropriate position in
110 the pixel when compressing from/decompressing to an RGB buffer.
112 Your application can check for the existence of these extensions at compile
115 #ifdef JCS_EXTENSIONS
117 At run time, attempting to use these extensions with a libjpeg implementation
118 that does not support them will result in a "Bogus input colorspace" error.
119 Applications can trap this error in order to test whether run-time support is
120 available for the colorspace extensions.
122 When using the RGBX, BGRX, XBGR, and XRGB colorspaces during decompression, the
123 X byte is undefined, and in order to ensure the best performance, libjpeg-turbo
124 can set that byte to whatever value it wishes. If an application expects the X
125 byte to be used as an alpha channel, then it should specify `JCS_EXT_RGBA`,
126 `JCS_EXT_BGRA`, `JCS_EXT_ABGR`, or `JCS_EXT_ARGB`. When these colorspace
127 constants are used, the X byte is guaranteed to be 0xFF, which is interpreted
130 Your application can check for the existence of the alpha channel colorspace
131 extensions at compile time with:
133 #ifdef JCS_ALPHA_EXTENSIONS
135 [jcstest.c](jcstest.c), located in the libjpeg-turbo source tree, demonstrates
136 how to check for the existence of the colorspace extensions at compile time and
139 libjpeg v7 and v8 API/ABI Emulation
140 -----------------------------------
142 With libjpeg v7 and v8, new features were added that necessitated extending the
143 compression and decompression structures. Unfortunately, due to the exposed
144 nature of those structures, extending them also necessitated breaking backward
145 ABI compatibility with previous libjpeg releases. Thus, programs that were
146 built to use libjpeg v7 or v8 did not work with libjpeg-turbo, since it is
147 based on the libjpeg v6b code base. Although libjpeg v7 and v8 are not
148 as widely used as v6b, enough programs (including a few Linux distros) made
149 the switch that there was a demand to emulate the libjpeg v7 and v8 ABIs
150 in libjpeg-turbo. It should be noted, however, that this feature was added
151 primarily so that applications that had already been compiled to use libjpeg
152 v7+ could take advantage of accelerated baseline JPEG encoding/decoding
153 without recompiling. libjpeg-turbo does not claim to support all of the
154 libjpeg v7+ features, nor to produce identical output to libjpeg v7+ in all
157 By passing an argument of `-DWITH_JPEG7=1` or `-DWITH_JPEG8=1` to `cmake`, you
158 can build a version of libjpeg-turbo that emulates the libjpeg v7 or v8 ABI, so
159 that programs that are built against libjpeg v7 or v8 can be run with
160 libjpeg-turbo. The following section describes which libjpeg v7+ features are
161 supported and which aren't.
163 ### Support for libjpeg v7 and v8 Features
167 - **libjpeg API: IDCT scaling extensions in decompressor**<br>
168 libjpeg-turbo supports IDCT scaling with scaling factors of 1/8, 1/4, 3/8,
169 1/2, 5/8, 3/4, 7/8, 9/8, 5/4, 11/8, 3/2, 13/8, 7/4, 15/8, and 2/1 (only 1/4
170 and 1/2 are SIMD-accelerated.)
172 - **libjpeg API: Arithmetic coding**
174 - **libjpeg API: In-memory source and destination managers**<br>
177 - **cjpeg: Separate quality settings for luminance and chrominance**<br>
178 Note that the libpjeg v7+ API was extended to accommodate this feature only
179 for convenience purposes. It has always been possible to implement this
180 feature with libjpeg v6b (see rdswitch.c for an example.)
182 - **cjpeg: 32-bit BMP support**
184 - **cjpeg: `-rgb` option**
186 - **jpegtran: Lossless cropping**
188 - **jpegtran: `-perfect` option**
190 - **jpegtran: Forcing width/height when performing lossless crop**
192 - **rdjpgcom: `-raw` option**
194 - **rdjpgcom: Locale awareness**
199 NOTE: As of this writing, extensive research has been conducted into the
200 usefulness of DCT scaling as a means of data reduction and SmartScale as a
201 means of quality improvement. Readers are invited to peruse the research at
202 <http://www.libjpeg-turbo.org/About/SmartScale> and draw their own conclusions,
203 but it is the general belief of our project that these features have not
204 demonstrated sufficient usefulness to justify inclusion in libjpeg-turbo.
206 - **libjpeg API: DCT scaling in compressor**<br>
207 `cinfo.scale_num` and `cinfo.scale_denom` are silently ignored.
208 There is no technical reason why DCT scaling could not be supported when
209 emulating the libjpeg v7+ API/ABI, but without the SmartScale extension (see
210 below), only scaling factors of 1/2, 8/15, 4/7, 8/13, 2/3, 8/11, 4/5, and
211 8/9 would be available, which is of limited usefulness.
213 - **libjpeg API: SmartScale**<br>
214 `cinfo.block_size` is silently ignored.
215 SmartScale is an extension to the JPEG format that allows for DCT block
216 sizes other than 8x8. Providing support for this new format would be
217 feasible (particularly without full acceleration.) However, until/unless
218 the format becomes either an official industry standard or, at minimum, an
219 accepted solution in the community, we are hesitant to implement it, as
220 there is no sense of whether or how it might change in the future. It is
221 our belief that SmartScale has not demonstrated sufficient usefulness as a
222 lossless format nor as a means of quality enhancement, and thus our primary
223 interest in providing this feature would be as a means of supporting
224 additional DCT scaling factors.
226 - **libjpeg API: Fancy downsampling in compressor**<br>
227 `cinfo.do_fancy_downsampling` is silently ignored.
228 This requires the DCT scaling feature, which is not supported.
230 - **jpegtran: Scaling**<br>
231 This requires both the DCT scaling and SmartScale features, which are not
234 - **Lossless RGB JPEG files**<br>
235 This requires the SmartScale feature, which is not supported.
237 ### What About libjpeg v9?
239 libjpeg v9 introduced yet another field to the JPEG compression structure
240 (`color_transform`), thus making the ABI backward incompatible with that of
241 libjpeg v8. This new field was introduced solely for the purpose of supporting
242 lossless SmartScale encoding. Furthermore, there was actually no reason to
243 extend the API in this manner, as the color transform could have just as easily
244 been activated by way of a new JPEG colorspace constant, thus preserving
245 backward ABI compatibility.
247 Our research (see link above) has shown that lossless SmartScale does not
248 generally accomplish anything that can't already be accomplished better with
249 existing, standard lossless formats. Therefore, at this time it is our belief
250 that there is not sufficient technical justification for software projects to
251 upgrade from libjpeg v8 to libjpeg v9, and thus there is not sufficient
252 technical justification for us to emulate the libjpeg v9 ABI.
254 In-Memory Source/Destination Managers
255 -------------------------------------
257 By default, libjpeg-turbo 1.3 and later includes the `jpeg_mem_src()` and
258 `jpeg_mem_dest()` functions, even when not emulating the libjpeg v8 API/ABI.
259 Previously, it was necessary to build libjpeg-turbo from source with libjpeg v8
260 API/ABI emulation in order to use the in-memory source/destination managers,
261 but several projects requested that those functions be included when emulating
262 the libjpeg v6b API/ABI as well. This allows the use of those functions by
263 programs that need them, without breaking ABI compatibility for programs that
264 don't, and it allows those functions to be provided in the "official"
265 libjpeg-turbo binaries.
267 Note that, on most Un*x systems, the dynamic linker will not look for a
268 function in a library until that function is actually used. Thus, if a program
269 is built against libjpeg-turbo 1.3+ and uses `jpeg_mem_src()` or
270 `jpeg_mem_dest()`, that program will not fail if run against an older version
271 of libjpeg-turbo or against libjpeg v7- until the program actually tries to
272 call `jpeg_mem_src()` or `jpeg_mem_dest()`. Such is not the case on Windows.
273 If a program is built against the libjpeg-turbo 1.3+ DLL and uses
274 `jpeg_mem_src()` or `jpeg_mem_dest()`, then it must use the libjpeg-turbo 1.3+
277 Both cjpeg and djpeg have been extended to allow testing the in-memory
278 source/destination manager functions. See their respective man pages for more
282 Mathematical Compatibility
283 ==========================
285 For the most part, libjpeg-turbo should produce identical output to libjpeg
286 v6b. There are two exceptions:
288 1. When decompressing a JPEG image that uses 4:4:0 chrominance subsampling, the
289 outputs of libjpeg v6b and libjpeg-turbo can differ because libjpeg-turbo
290 implements a "fancy" (smooth) 4:4:0 upsampling algorithm and libjpeg did not.
292 2. When using the floating point DCT/IDCT, the outputs of libjpeg v6b and
293 libjpeg-turbo can differ for the following reasons:
295 - The SSE/SSE2 floating point DCT implementation in libjpeg-turbo is ever
296 so slightly more accurate than the implementation in libjpeg v6b, but not
297 by any amount perceptible to human vision (generally in the range of 0.01
298 to 0.08 dB gain in PNSR.)
300 - When not using the SIMD extensions, libjpeg-turbo uses the more accurate
301 (and slightly faster) floating point IDCT algorithm introduced in libjpeg
302 v8a as opposed to the algorithm used in libjpeg v6b. It should be noted,
303 however, that this algorithm basically brings the accuracy of the
304 floating point IDCT in line with the accuracy of the accurate integer
305 IDCT. The floating point DCT/IDCT algorithms are mainly a legacy
306 feature, and they do not produce significantly more accuracy than the
307 accurate integer algorithms. (To put numbers on this, the typical
308 difference in PNSR between the two algorithms is less than 0.10 dB,
309 whereas changing the quality level by 1 in the upper range of the quality
310 scale is typically more like a 1.0 dB difference.)
312 - If the floating point algorithms in libjpeg-turbo are not implemented
313 using SIMD instructions on a particular platform, then the accuracy of
314 the floating point DCT/IDCT can depend on the compiler settings.
316 While libjpeg-turbo does emulate the libjpeg v8 API/ABI, under the hood it is
317 still using the same algorithms as libjpeg v6b, so there are several specific
318 cases in which libjpeg-turbo cannot be expected to produce the same output as
321 - When decompressing using scaling factors of 1/2 and 1/4, because libjpeg v8
322 implements those scaling algorithms differently than libjpeg v6b does, and
323 libjpeg-turbo's SIMD extensions are based on the libjpeg v6b behavior.
325 - When using chrominance subsampling, because libjpeg v8 implements this
326 with its DCT/IDCT scaling algorithms rather than with a separate
327 downsampling/upsampling algorithm. In our testing, the subsampled/upsampled
328 output of libjpeg v8 is less accurate than that of libjpeg v6b for this
331 - When decompressing using a scaling factor > 1 and merged (AKA "non-fancy" or
332 "non-smooth") chrominance upsampling, because libjpeg v8 does not support
333 merged upsampling with scaling factors > 1.
342 The optimized Huffman decoder in libjpeg-turbo does not handle restart markers
343 in a way that makes the rest of the libjpeg infrastructure happy, so it is
344 necessary to use the slow Huffman decoder when decompressing a JPEG image that
345 has restart markers. This can cause the decompression performance to drop by
346 as much as 20%, but the performance will still be much greater than that of
347 libjpeg. Many consumer packages, such as Photoshop, use restart markers when
348 generating JPEG images, so images generated by those programs will experience
351 Fast Integer Forward DCT at High Quality Levels
352 -----------------------------------------------
354 The algorithm used by the SIMD-accelerated quantization function cannot produce
355 correct results whenever the fast integer forward DCT is used along with a JPEG
356 quality of 98-100. Thus, libjpeg-turbo must use the non-SIMD quantization
357 function in those cases. This causes performance to drop by as much as 40%.
358 It is therefore strongly advised that you use the accurate integer forward DCT
359 whenever encoding images with a JPEG quality of 98 or higher.
362 Memory Debugger Pitfalls
363 ========================
365 Valgrind and Memory Sanitizer (MSan) can generate false positives
366 (specifically, incorrect reports of uninitialized memory accesses) when used
367 with libjpeg-turbo's SIMD extensions. It is generally recommended that the
368 SIMD extensions be disabled, either by passing an argument of `-DWITH_SIMD=0`
369 to `cmake` when configuring the build or by setting the environment variable
370 `JSIMD_FORCENONE` to `1` at run time, when testing libjpeg-turbo with Valgrind,
371 MSan, or other memory debuggers.