1 /*M///////////////////////////////////////////////////////////////////////////////////////
3 // IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
5 // By downloading, copying, installing or using the software you agree to this license.
6 // If you do not agree to this license, do not download, install,
7 // copy or use the software.
11 // For Open Source Computer Vision Library
13 // Copyright (C) 2010-2012, Institute Of Software Chinese Academy Of Science, all rights reserved.
14 // Copyright (C) 2010-2012, Advanced Micro Devices, Inc., all rights reserved.
15 // Copyright (C) 2010-2012, Multicoreware, Inc., all rights reserved.
16 // Third party copyrights are property of their respective owners.
18 // Redistribution and use in source and binary forms, with or without modification,
19 // are permitted provided that the following conditions are met:
21 // * Redistribution's of source code must retain the above copyright notice,
22 // this list of conditions and the following disclaimer.
24 // * Redistribution's in binary form must reproduce the above copyright notice,
25 // this list of conditions and the following disclaimer in the documentation
26 // and/or other materials provided with the distribution.
28 // * The name of the copyright holders may not be used to endorse or promote products
29 // derived from this software without specific prior written permission.
31 // This software is provided by the copyright holders and contributors "as is" and
32 // any express or implied warranties, including, but not limited to, the implied
33 // warranties of merchantability and fitness for a particular purpose are disclaimed.
34 // In no event shall the Intel Corporation or contributors be liable for any direct,
35 // indirect, incidental, special, exemplary, or consequential damages
36 // (including, but not limited to, procurement of substitute goods or services;
37 // loss of use, data, or profits; or business interruption) however caused
38 // and on any theory of liability, whether in contract, strict liability,
39 // or tort (including negligence or otherwise) arising in any way out of
40 // the use of this software, even if advised of the possibility of such damage.
44 #ifndef __OPENCV_OCL_HPP__
45 #define __OPENCV_OCL_HPP__
50 #include "opencv2/core.hpp"
51 #include "opencv2/imgproc.hpp"
52 #include "opencv2/objdetect.hpp"
53 #include "opencv2/ml.hpp"
61 CVCL_DEVICE_TYPE_DEFAULT = (1 << 0),
62 CVCL_DEVICE_TYPE_CPU = (1 << 1),
63 CVCL_DEVICE_TYPE_GPU = (1 << 2),
64 CVCL_DEVICE_TYPE_ACCELERATOR = (1 << 3),
65 //CVCL_DEVICE_TYPE_CUSTOM = (1 << 4)
66 CVCL_DEVICE_TYPE_ALL = 0xFFFFFFFF
78 DEVICE_MEM_DEFAULT = 0,
79 DEVICE_MEM_AHP, //alloc host pointer
80 DEVICE_MEM_UHP, //use host pointer
81 DEVICE_MEM_CHP, //copy host pointer
82 DEVICE_MEM_PM //persistent memory
85 // these classes contain OpenCL runtime information
92 int _id; // reserved, don't use it
94 DeviceType deviceType;
95 std::string deviceProfile;
96 std::string deviceVersion;
97 std::string deviceName;
98 std::string deviceVendor;
100 std::string deviceDriverVersion;
101 std::string deviceExtensions;
103 size_t maxWorkGroupSize;
104 std::vector<size_t> maxWorkItemSizes;
106 size_t localMemorySize;
107 size_t maxMemAllocSize;
109 int deviceVersionMajor;
110 int deviceVersionMinor;
112 bool haveDoubleSupport;
113 bool isUnifiedMemory; // 1 means integrated GPU, otherwise this value is 0
116 std::string compilationExtraOptions;
118 const PlatformInfo* platform;
125 int _id; // reserved, don't use it
127 std::string platformProfile;
128 std::string platformVersion;
129 std::string platformName;
130 std::string platformVendor;
131 std::string platformExtensons;
133 int platformVersionMajor;
134 int platformVersionMinor;
136 std::vector<const DeviceInfo*> devices;
141 //////////////////////////////// Initialization & Info ////////////////////////
142 typedef std::vector<const PlatformInfo*> PlatformsInfo;
144 CV_EXPORTS int getOpenCLPlatforms(PlatformsInfo& platforms);
146 typedef std::vector<const DeviceInfo*> DevicesInfo;
148 CV_EXPORTS int getOpenCLDevices(DevicesInfo& devices, int deviceType = CVCL_DEVICE_TYPE_GPU,
149 const PlatformInfo* platform = NULL);
151 // set device you want to use
152 CV_EXPORTS void setDevice(const DeviceInfo* info);
156 FEATURE_CL_DOUBLE = 1,
157 FEATURE_CL_UNIFIED_MEM,
159 FEATURE_CL_INTEL_DEVICE
162 // Represents OpenCL context, interface
163 class CV_EXPORTS Context
169 static Context *getContext();
171 bool supportsFeature(FEATURE_TYPE featureType) const;
172 const DeviceInfo& getDeviceInfo() const;
174 const void* getOpenCLContextPtr() const;
175 const void* getOpenCLCommandQueuePtr() const;
176 const void* getOpenCLDeviceIDPtr() const;
179 inline const void *getClContextPtr()
181 return Context::getContext()->getOpenCLContextPtr();
184 inline const void *getClCommandQueuePtr()
186 return Context::getContext()->getOpenCLCommandQueuePtr();
189 CV_EXPORTS bool supportsFeature(FEATURE_TYPE featureType);
191 CV_EXPORTS void finish();
193 enum BINARY_CACHE_MODE
195 CACHE_NONE = 0, // do not cache OpenCL binary
196 CACHE_DEBUG = 0x1 << 0, // cache OpenCL binary when built in debug mode
197 CACHE_RELEASE = 0x1 << 1, // default behavior, only cache when built in release mode
198 CACHE_ALL = CACHE_DEBUG | CACHE_RELEASE, // cache opencl binary
200 //! Enable or disable OpenCL program binary caching onto local disk
201 // After a program (*.cl files in opencl/ folder) is built at runtime, we allow the
202 // compiled OpenCL program to be cached to the path automatically as "path/*.clb"
203 // binary file, which will be reused when the OpenCV executable is started again.
205 // This feature is enabled by default.
206 CV_EXPORTS void setBinaryDiskCache(int mode = CACHE_RELEASE, cv::String path = "./");
208 //! set where binary cache to be saved to
209 CV_EXPORTS void setBinaryPath(const char *path);
214 const char* programStr;
215 const char* programHash;
217 // Cache in memory by name (should be unique). Caching on disk disabled.
218 inline ProgramSource(const char* _name, const char* _programStr)
219 : name(_name), programStr(_programStr), programHash(NULL)
223 // Cache in memory by name (should be unique). Caching on disk uses programHash mark.
224 inline ProgramSource(const char* _name, const char* _programStr, const char* _programHash)
225 : name(_name), programStr(_programStr), programHash(_programHash)
230 //! Calls OpenCL kernel. Pass globalThreads = NULL, and cleanUp = true, to finally clean-up without executing.
231 //! Deprecated, will be replaced
232 CV_EXPORTS void openCLExecuteKernelInterop(Context *clCxt,
233 const cv::ocl::ProgramSource& source, String kernelName,
234 size_t globalThreads[3], size_t localThreads[3],
235 std::vector< std::pair<size_t, const void *> > &args,
236 int channels, int depth, const char *build_options);
238 class CV_EXPORTS oclMatExpr;
239 //////////////////////////////// oclMat ////////////////////////////////
240 class CV_EXPORTS oclMat
243 //! default constructor
245 //! constructs oclMatrix of the specified size and type (_type is CV_8UC1, CV_64FC3, CV_32SC(12) etc.)
246 oclMat(int rows, int cols, int type);
247 oclMat(Size size, int type);
248 //! constucts oclMatrix and fills it with the specified value _s.
249 oclMat(int rows, int cols, int type, const Scalar &s);
250 oclMat(Size size, int type, const Scalar &s);
252 oclMat(const oclMat &m);
254 //! constructor for oclMatrix headers pointing to user-allocated data
255 oclMat(int rows, int cols, int type, void *data, size_t step = Mat::AUTO_STEP);
256 oclMat(Size size, int type, void *data, size_t step = Mat::AUTO_STEP);
258 //! creates a matrix header for a part of the bigger matrix
259 oclMat(const oclMat &m, const Range &rowRange, const Range &colRange);
260 oclMat(const oclMat &m, const Rect &roi);
262 //! builds oclMat from Mat. Perfom blocking upload to device.
263 explicit oclMat (const Mat &m);
265 //! destructor - calls release()
268 //! assignment operators
269 oclMat &operator = (const oclMat &m);
270 //! assignment operator. Perfom blocking upload to device.
271 oclMat &operator = (const Mat &m);
272 oclMat &operator = (const oclMatExpr& expr);
274 //! pefroms blocking upload data to oclMat.
275 void upload(const cv::Mat &m);
278 //! downloads data from device to host memory. Blocking calls.
279 operator Mat() const;
280 void download(cv::Mat &m) const;
282 //! convert to _InputArray
283 operator _InputArray();
285 //! convert to _OutputArray
286 operator _OutputArray();
288 //! returns a new oclMatrix header for the specified row
289 oclMat row(int y) const;
290 //! returns a new oclMatrix header for the specified column
291 oclMat col(int x) const;
292 //! ... for the specified row span
293 oclMat rowRange(int startrow, int endrow) const;
294 oclMat rowRange(const Range &r) const;
295 //! ... for the specified column span
296 oclMat colRange(int startcol, int endcol) const;
297 oclMat colRange(const Range &r) const;
299 //! returns deep copy of the oclMatrix, i.e. the data is copied
300 oclMat clone() const;
302 //! copies those oclMatrix elements to "m" that are marked with non-zero mask elements.
303 // It calls m.create(this->size(), this->type()).
304 // It supports any data type
305 void copyTo( oclMat &m, const oclMat &mask = oclMat()) const;
307 //! converts oclMatrix to another datatype with optional scalng. See cvConvertScale.
308 void convertTo( oclMat &m, int rtype, double alpha = 1, double beta = 0 ) const;
310 void assignTo( oclMat &m, int type = -1 ) const;
312 //! sets every oclMatrix element to s
313 oclMat& operator = (const Scalar &s);
314 //! sets some of the oclMatrix elements to s, according to the mask
315 oclMat& setTo(const Scalar &s, const oclMat &mask = oclMat());
316 //! creates alternative oclMatrix header for the same data, with different
317 // number of channels and/or different number of rows. see cvReshape.
318 oclMat reshape(int cn, int rows = 0) const;
320 //! allocates new oclMatrix data unless the oclMatrix already has specified size and type.
321 // previous data is unreferenced if needed.
322 void create(int rows, int cols, int type);
323 void create(Size size, int type);
325 //! allocates new oclMatrix with specified device memory type.
326 void createEx(int rows, int cols, int type,
327 DevMemRW rw_type, DevMemType mem_type);
328 void createEx(Size size, int type, DevMemRW rw_type,
329 DevMemType mem_type);
331 //! decreases reference counter;
332 // deallocate the data when reference counter reaches 0.
335 //! swaps with other smart pointer
336 void swap(oclMat &mat);
338 //! locates oclMatrix header within a parent oclMatrix. See below
339 void locateROI( Size &wholeSize, Point &ofs ) const;
340 //! moves/resizes the current oclMatrix ROI inside the parent oclMatrix.
341 oclMat& adjustROI( int dtop, int dbottom, int dleft, int dright );
342 //! extracts a rectangular sub-oclMatrix
343 // (this is a generalized form of row, rowRange etc.)
344 oclMat operator()( Range rowRange, Range colRange ) const;
345 oclMat operator()( const Rect &roi ) const;
347 oclMat& operator+=( const oclMat& m );
348 oclMat& operator-=( const oclMat& m );
349 oclMat& operator*=( const oclMat& m );
350 oclMat& operator/=( const oclMat& m );
352 //! returns true if the oclMatrix data is continuous
353 // (i.e. when there are no gaps between successive rows).
354 // similar to CV_IS_oclMat_CONT(cvoclMat->type)
355 bool isContinuous() const;
356 //! returns element size in bytes,
357 // similar to CV_ELEM_SIZE(cvMat->type)
358 size_t elemSize() const;
359 //! returns the size of element channel in bytes.
360 size_t elemSize1() const;
361 //! returns element type, similar to CV_MAT_TYPE(cvMat->type)
363 //! returns element type, i.e. 8UC3 returns 8UC4 because in ocl
364 //! 3 channels element actually use 4 channel space
366 //! returns element type, similar to CV_MAT_DEPTH(cvMat->type)
368 //! returns element type, similar to CV_MAT_CN(cvMat->type)
369 int channels() const;
370 //! returns element type, return 4 for 3 channels element,
371 //!becuase 3 channels element actually use 4 channel space
372 int oclchannels() const;
373 //! returns step/elemSize1()
374 size_t step1() const;
375 //! returns oclMatrix size:
376 // width == number of columns, height == number of rows
378 //! returns true if oclMatrix data is NULL
381 //! returns pointer to y-th row
382 uchar* ptr(int y = 0);
383 const uchar *ptr(int y = 0) const;
385 //! template version of the above method
386 template<typename _Tp> _Tp *ptr(int y = 0);
387 template<typename _Tp> const _Tp *ptr(int y = 0) const;
389 //! matrix transposition
392 /*! includes several bit-fields:
393 - the magic signature
399 //! the number of rows and columns
401 //! a distance between successive rows in bytes; includes the gap if any
403 //! pointer to the data(OCL memory object)
406 //! pointer to the reference counter;
407 // when oclMatrix points to user-allocated data, the pointer is NULL
410 //! helper fields used in locateROI and adjustROI
411 //datastart and dataend are not used in current version
415 //! OpenCL context associated with the oclMat object.
416 Context *clCxt; // TODO clCtx
417 //add offset for handle ROI, calculated in byte
419 //add wholerows and wholecols for the whole matrix, datastart and dataend are no longer used
424 // convert InputArray/OutputArray to oclMat references
425 CV_EXPORTS oclMat& getOclMatRef(InputArray src);
426 CV_EXPORTS oclMat& getOclMatRef(OutputArray src);
428 ///////////////////// mat split and merge /////////////////////////////////
429 //! Compose a multi-channel array from several single-channel arrays
431 CV_EXPORTS void merge(const oclMat *src, size_t n, oclMat &dst);
432 CV_EXPORTS void merge(const std::vector<oclMat> &src, oclMat &dst);
434 //! Divides multi-channel array into several single-channel arrays
436 CV_EXPORTS void split(const oclMat &src, oclMat *dst);
437 CV_EXPORTS void split(const oclMat &src, std::vector<oclMat> &dst);
439 ////////////////////////////// Arithmetics ///////////////////////////////////
441 //! adds one matrix to another with scale (dst = src1 * alpha + src2 * beta + gama)
442 // supports all data types
443 CV_EXPORTS void addWeighted(const oclMat &src1, double alpha, const oclMat &src2, double beta, double gama, oclMat &dst);
445 //! adds one matrix to another (dst = src1 + src2)
446 // supports all data types
447 CV_EXPORTS void add(const oclMat &src1, const oclMat &src2, oclMat &dst, const oclMat &mask = oclMat());
448 //! adds scalar to a matrix (dst = src1 + s)
449 // supports all data types
450 CV_EXPORTS void add(const oclMat &src1, const Scalar &s, oclMat &dst, const oclMat &mask = oclMat());
452 //! subtracts one matrix from another (dst = src1 - src2)
453 // supports all data types
454 CV_EXPORTS void subtract(const oclMat &src1, const oclMat &src2, oclMat &dst, const oclMat &mask = oclMat());
455 //! subtracts scalar from a matrix (dst = src1 - s)
456 // supports all data types
457 CV_EXPORTS void subtract(const oclMat &src1, const Scalar &s, oclMat &dst, const oclMat &mask = oclMat());
459 //! computes element-wise product of the two arrays (dst = src1 * scale * src2)
460 // supports all data types
461 CV_EXPORTS void multiply(const oclMat &src1, const oclMat &src2, oclMat &dst, double scale = 1);
462 //! multiplies matrix to a number (dst = scalar * src)
463 // supports all data types
464 CV_EXPORTS void multiply(double scalar, const oclMat &src, oclMat &dst);
466 //! computes element-wise quotient of the two arrays (dst = src1 * scale / src2)
467 // supports all data types
468 CV_EXPORTS void divide(const oclMat &src1, const oclMat &src2, oclMat &dst, double scale = 1);
469 //! computes element-wise quotient of the two arrays (dst = scale / src)
470 // supports all data types
471 CV_EXPORTS void divide(double scale, const oclMat &src1, oclMat &dst);
473 //! computes element-wise minimum of the two arrays (dst = min(src1, src2))
474 // supports all data types
475 CV_EXPORTS void min(const oclMat &src1, const oclMat &src2, oclMat &dst);
477 //! computes element-wise maximum of the two arrays (dst = max(src1, src2))
478 // supports all data types
479 CV_EXPORTS void max(const oclMat &src1, const oclMat &src2, oclMat &dst);
481 //! compares elements of two arrays (dst = src1 <cmpop> src2)
482 // supports all data types
483 CV_EXPORTS void compare(const oclMat &src1, const oclMat &src2, oclMat &dst, int cmpop);
485 //! transposes the matrix
486 // supports all data types
487 CV_EXPORTS void transpose(const oclMat &src, oclMat &dst);
489 //! computes element-wise absolute values of an array (dst = abs(src))
490 // supports all data types
491 CV_EXPORTS void abs(const oclMat &src, oclMat &dst);
493 //! computes element-wise absolute difference of two arrays (dst = abs(src1 - src2))
494 // supports all data types
495 CV_EXPORTS void absdiff(const oclMat &src1, const oclMat &src2, oclMat &dst);
496 //! computes element-wise absolute difference of array and scalar (dst = abs(src1 - s))
497 // supports all data types
498 CV_EXPORTS void absdiff(const oclMat &src1, const Scalar &s, oclMat &dst);
500 //! computes mean value and standard deviation of all or selected array elements
501 // supports all data types
502 CV_EXPORTS void meanStdDev(const oclMat &mtx, Scalar &mean, Scalar &stddev);
504 //! computes norm of array
505 // supports NORM_INF, NORM_L1, NORM_L2
506 // supports all data types
507 CV_EXPORTS double norm(const oclMat &src1, int normType = NORM_L2);
509 //! computes norm of the difference between two arrays
510 // supports NORM_INF, NORM_L1, NORM_L2
511 // supports all data types
512 CV_EXPORTS double norm(const oclMat &src1, const oclMat &src2, int normType = NORM_L2);
514 //! reverses the order of the rows, columns or both in a matrix
515 // supports all types
516 CV_EXPORTS void flip(const oclMat &src, oclMat &dst, int flipCode);
518 //! computes sum of array elements
520 CV_EXPORTS Scalar sum(const oclMat &m);
521 CV_EXPORTS Scalar absSum(const oclMat &m);
522 CV_EXPORTS Scalar sqrSum(const oclMat &m);
524 //! finds global minimum and maximum array elements and returns their values
525 // support all C1 types
526 CV_EXPORTS void minMax(const oclMat &src, double *minVal, double *maxVal = 0, const oclMat &mask = oclMat());
528 //! finds global minimum and maximum array elements and returns their values with locations
529 // support all C1 types
530 CV_EXPORTS void minMaxLoc(const oclMat &src, double *minVal, double *maxVal = 0, Point *minLoc = 0, Point *maxLoc = 0,
531 const oclMat &mask = oclMat());
533 //! counts non-zero array elements
535 CV_EXPORTS int countNonZero(const oclMat &src);
537 //! transforms 8-bit unsigned integers using lookup table: dst(i)=lut(src(i))
538 // destination array will have the depth type as lut and the same channels number as source
539 //It supports 8UC1 8UC4 only
540 CV_EXPORTS void LUT(const oclMat &src, const oclMat &lut, oclMat &dst);
542 //! only 8UC1 and 256 bins is supported now
543 CV_EXPORTS void calcHist(const oclMat &mat_src, oclMat &mat_hist);
544 //! only 8UC1 and 256 bins is supported now
545 CV_EXPORTS void equalizeHist(const oclMat &mat_src, oclMat &mat_dst);
547 //! only 8UC1 is supported now
548 CV_EXPORTS Ptr<cv::CLAHE> createCLAHE(double clipLimit = 40.0, Size tileGridSize = Size(8, 8));
551 // supports 8UC1 8UC4
552 CV_EXPORTS void bilateralFilter(const oclMat& src, oclMat& dst, int d, double sigmaColor, double sigmaSpace, int borderType=BORDER_DEFAULT);
554 //! Applies an adaptive bilateral filter to the input image
555 // Unlike the usual bilateral filter that uses fixed value for sigmaColor,
556 // the adaptive version calculates the local variance in he ksize neighborhood
557 // and use this as sigmaColor, for the value filtering. However, the local standard deviation is
558 // clamped to the maxSigmaColor.
559 // supports 8UC1, 8UC3
560 CV_EXPORTS void adaptiveBilateralFilter(const oclMat& src, oclMat& dst, Size ksize, double sigmaSpace, double maxSigmaColor=20.0, Point anchor = Point(-1, -1), int borderType=BORDER_DEFAULT);
562 //! computes exponent of each matrix element (dst = e**src)
563 // supports only CV_32FC1, CV_64FC1 type
564 CV_EXPORTS void exp(const oclMat &src, oclMat &dst);
566 //! computes natural logarithm of absolute value of each matrix element: dst = log(abs(src))
567 // supports only CV_32FC1, CV_64FC1 type
568 CV_EXPORTS void log(const oclMat &src, oclMat &dst);
570 //! computes magnitude of each (x(i), y(i)) vector
571 // supports only CV_32F, CV_64F type
572 CV_EXPORTS void magnitude(const oclMat &x, const oclMat &y, oclMat &magnitude);
574 //! computes angle (angle(i)) of each (x(i), y(i)) vector
575 // supports only CV_32F, CV_64F type
576 CV_EXPORTS void phase(const oclMat &x, const oclMat &y, oclMat &angle, bool angleInDegrees = false);
578 //! the function raises every element of tne input array to p
579 // support only CV_32F, CV_64F type
580 CV_EXPORTS void pow(const oclMat &x, double p, oclMat &y);
582 //! converts Cartesian coordinates to polar
583 // supports only CV_32F CV_64F type
584 CV_EXPORTS void cartToPolar(const oclMat &x, const oclMat &y, oclMat &magnitude, oclMat &angle, bool angleInDegrees = false);
586 //! converts polar coordinates to Cartesian
587 // supports only CV_32F CV_64F type
588 CV_EXPORTS void polarToCart(const oclMat &magnitude, const oclMat &angle, oclMat &x, oclMat &y, bool angleInDegrees = false);
590 //! perfroms per-elements bit-wise inversion
591 // supports all types
592 CV_EXPORTS void bitwise_not(const oclMat &src, oclMat &dst);
594 //! calculates per-element bit-wise disjunction of two arrays
595 // supports all types
596 CV_EXPORTS void bitwise_or(const oclMat &src1, const oclMat &src2, oclMat &dst, const oclMat &mask = oclMat());
597 CV_EXPORTS void bitwise_or(const oclMat &src1, const Scalar &s, oclMat &dst, const oclMat &mask = oclMat());
599 //! calculates per-element bit-wise conjunction of two arrays
600 // supports all types
601 CV_EXPORTS void bitwise_and(const oclMat &src1, const oclMat &src2, oclMat &dst, const oclMat &mask = oclMat());
602 CV_EXPORTS void bitwise_and(const oclMat &src1, const Scalar &s, oclMat &dst, const oclMat &mask = oclMat());
604 //! calculates per-element bit-wise "exclusive or" operation
605 // supports all types
606 CV_EXPORTS void bitwise_xor(const oclMat &src1, const oclMat &src2, oclMat &dst, const oclMat &mask = oclMat());
607 CV_EXPORTS void bitwise_xor(const oclMat &src1, const Scalar &s, oclMat &dst, const oclMat &mask = oclMat());
609 //! Logical operators
610 CV_EXPORTS oclMat operator ~ (const oclMat &);
611 CV_EXPORTS oclMat operator | (const oclMat &, const oclMat &);
612 CV_EXPORTS oclMat operator & (const oclMat &, const oclMat &);
613 CV_EXPORTS oclMat operator ^ (const oclMat &, const oclMat &);
616 //! Mathematics operators
617 CV_EXPORTS oclMatExpr operator + (const oclMat &src1, const oclMat &src2);
618 CV_EXPORTS oclMatExpr operator - (const oclMat &src1, const oclMat &src2);
619 CV_EXPORTS oclMatExpr operator * (const oclMat &src1, const oclMat &src2);
620 CV_EXPORTS oclMatExpr operator / (const oclMat &src1, const oclMat &src2);
622 struct CV_EXPORTS ConvolveBuf
626 Size user_block_size;
629 oclMat image_spect, templ_spect, result_spect;
630 oclMat image_block, templ_block, result_data;
632 void create(Size image_size, Size templ_size);
633 static Size estimateBlockSize(Size result_size, Size templ_size);
636 //! computes convolution of two images, may use discrete Fourier transform
637 // support only CV_32FC1 type
638 CV_EXPORTS void convolve(const oclMat &image, const oclMat &temp1, oclMat &result, bool ccorr = false);
639 CV_EXPORTS void convolve(const oclMat &image, const oclMat &temp1, oclMat &result, bool ccorr, ConvolveBuf& buf);
641 //! Performs a per-element multiplication of two Fourier spectrums.
642 //! Only full (not packed) CV_32FC2 complex spectrums in the interleaved format are supported for now.
643 //! support only CV_32FC2 type
644 CV_EXPORTS void mulSpectrums(const oclMat &a, const oclMat &b, oclMat &c, int flags, float scale, bool conjB = false);
646 CV_EXPORTS void cvtColor(const oclMat &src, oclMat &dst, int code, int dcn = 0);
648 //! initializes a scaled identity matrix
649 CV_EXPORTS void setIdentity(oclMat& src, const Scalar & val = Scalar(1));
651 //! fills the output array with repeated copies of the input array
652 CV_EXPORTS void repeat(const oclMat & src, int ny, int nx, oclMat & dst);
654 //////////////////////////////// Filter Engine ////////////////////////////////
657 The Base Class for 1D or Row-wise Filters
659 This is the base class for linear or non-linear filters that process 1D data.
660 In particular, such filters are used for the "horizontal" filtering parts in separable filters.
662 class CV_EXPORTS BaseRowFilter_GPU
665 BaseRowFilter_GPU(int ksize_, int anchor_, int bordertype_) : ksize(ksize_), anchor(anchor_), bordertype(bordertype_) {}
666 virtual ~BaseRowFilter_GPU() {}
667 virtual void operator()(const oclMat &src, oclMat &dst) = 0;
668 int ksize, anchor, bordertype;
672 The Base Class for Column-wise Filters
674 This is the base class for linear or non-linear filters that process columns of 2D arrays.
675 Such filters are used for the "vertical" filtering parts in separable filters.
677 class CV_EXPORTS BaseColumnFilter_GPU
680 BaseColumnFilter_GPU(int ksize_, int anchor_, int bordertype_) : ksize(ksize_), anchor(anchor_), bordertype(bordertype_) {}
681 virtual ~BaseColumnFilter_GPU() {}
682 virtual void operator()(const oclMat &src, oclMat &dst) = 0;
683 int ksize, anchor, bordertype;
687 The Base Class for Non-Separable 2D Filters.
689 This is the base class for linear or non-linear 2D filters.
691 class CV_EXPORTS BaseFilter_GPU
694 BaseFilter_GPU(const Size &ksize_, const Point &anchor_, const int &borderType_)
695 : ksize(ksize_), anchor(anchor_), borderType(borderType_) {}
696 virtual ~BaseFilter_GPU() {}
697 virtual void operator()(const oclMat &src, oclMat &dst) = 0;
704 The Base Class for Filter Engine.
706 The class can be used to apply an arbitrary filtering operation to an image.
707 It contains all the necessary intermediate buffers.
709 class CV_EXPORTS FilterEngine_GPU
712 virtual ~FilterEngine_GPU() {}
714 virtual void apply(const oclMat &src, oclMat &dst, Rect roi = Rect(0, 0, -1, -1)) = 0;
717 //! returns the non-separable filter engine with the specified filter
718 CV_EXPORTS Ptr<FilterEngine_GPU> createFilter2D_GPU(const Ptr<BaseFilter_GPU> filter2D);
720 //! returns the primitive row filter with the specified kernel
721 CV_EXPORTS Ptr<BaseRowFilter_GPU> getLinearRowFilter_GPU(int srcType, int bufType, const Mat &rowKernel,
722 int anchor = -1, int bordertype = BORDER_DEFAULT);
724 //! returns the primitive column filter with the specified kernel
725 CV_EXPORTS Ptr<BaseColumnFilter_GPU> getLinearColumnFilter_GPU(int bufType, int dstType, const Mat &columnKernel,
726 int anchor = -1, int bordertype = BORDER_DEFAULT, double delta = 0.0);
728 //! returns the separable linear filter engine
729 CV_EXPORTS Ptr<FilterEngine_GPU> createSeparableLinearFilter_GPU(int srcType, int dstType, const Mat &rowKernel,
730 const Mat &columnKernel, const Point &anchor = Point(-1, -1), double delta = 0.0, int bordertype = BORDER_DEFAULT);
732 //! returns the separable filter engine with the specified filters
733 CV_EXPORTS Ptr<FilterEngine_GPU> createSeparableFilter_GPU(const Ptr<BaseRowFilter_GPU> &rowFilter,
734 const Ptr<BaseColumnFilter_GPU> &columnFilter);
736 //! returns the Gaussian filter engine
737 CV_EXPORTS Ptr<FilterEngine_GPU> createGaussianFilter_GPU(int type, Size ksize, double sigma1, double sigma2 = 0, int bordertype = BORDER_DEFAULT);
739 //! returns filter engine for the generalized Sobel operator
740 CV_EXPORTS Ptr<FilterEngine_GPU> createDerivFilter_GPU( int srcType, int dstType, int dx, int dy, int ksize, int borderType = BORDER_DEFAULT );
742 //! applies Laplacian operator to the image
743 // supports only ksize = 1 and ksize = 3
744 CV_EXPORTS void Laplacian(const oclMat &src, oclMat &dst, int ddepth, int ksize = 1, double scale = 1,
745 double delta=0, int borderType=BORDER_DEFAULT);
747 //! returns 2D box filter
748 // dst type must be the same as source type
749 CV_EXPORTS Ptr<BaseFilter_GPU> getBoxFilter_GPU(int srcType, int dstType,
750 const Size &ksize, Point anchor = Point(-1, -1), int borderType = BORDER_DEFAULT);
752 //! returns box filter engine
753 CV_EXPORTS Ptr<FilterEngine_GPU> createBoxFilter_GPU(int srcType, int dstType, const Size &ksize,
754 const Point &anchor = Point(-1, -1), int borderType = BORDER_DEFAULT);
756 //! returns 2D filter with the specified kernel
757 // supports: dst type must be the same as source type
758 CV_EXPORTS Ptr<BaseFilter_GPU> getLinearFilter_GPU(int srcType, int dstType, const Mat &kernel, const Size &ksize,
759 const Point &anchor = Point(-1, -1), int borderType = BORDER_DEFAULT);
761 //! returns the non-separable linear filter engine
762 // supports: dst type must be the same as source type
763 CV_EXPORTS Ptr<FilterEngine_GPU> createLinearFilter_GPU(int srcType, int dstType, const Mat &kernel,
764 const Point &anchor = Point(-1, -1), int borderType = BORDER_DEFAULT);
766 //! smooths the image using the normalized box filter
767 CV_EXPORTS void boxFilter(const oclMat &src, oclMat &dst, int ddepth, Size ksize,
768 Point anchor = Point(-1, -1), int borderType = BORDER_DEFAULT);
770 //! returns 2D morphological filter
771 //! only MORPH_ERODE and MORPH_DILATE are supported
772 // supports CV_8UC1, CV_8UC4, CV_32FC1 and CV_32FC4 types
773 // kernel must have CV_8UC1 type, one rows and cols == ksize.width * ksize.height
774 CV_EXPORTS Ptr<BaseFilter_GPU> getMorphologyFilter_GPU(int op, int type, const Mat &kernel, const Size &ksize,
775 Point anchor = Point(-1, -1));
777 //! returns morphological filter engine. Only MORPH_ERODE and MORPH_DILATE are supported.
778 CV_EXPORTS Ptr<FilterEngine_GPU> createMorphologyFilter_GPU(int op, int type, const Mat &kernel,
779 const Point &anchor = Point(-1, -1), int iterations = 1);
781 //! a synonym for normalized box filter
782 static inline void blur(const oclMat &src, oclMat &dst, Size ksize, Point anchor = Point(-1, -1),
783 int borderType = BORDER_CONSTANT)
785 boxFilter(src, dst, -1, ksize, anchor, borderType);
788 //! applies non-separable 2D linear filter to the image
789 CV_EXPORTS void filter2D(const oclMat &src, oclMat &dst, int ddepth, const Mat &kernel,
790 Point anchor = Point(-1, -1), double delta = 0.0, int borderType = BORDER_DEFAULT);
792 //! applies separable 2D linear filter to the image
793 CV_EXPORTS void sepFilter2D(const oclMat &src, oclMat &dst, int ddepth, const Mat &kernelX, const Mat &kernelY,
794 Point anchor = Point(-1, -1), double delta = 0.0, int bordertype = BORDER_DEFAULT);
796 //! applies generalized Sobel operator to the image
797 // dst.type must equalize src.type
798 // supports data type: CV_8UC1, CV_8UC4, CV_32FC1 and CV_32FC4
799 // supports border type: BORDER_CONSTANT, BORDER_REPLICATE, BORDER_REFLECT,BORDER_REFLECT_101
800 CV_EXPORTS void Sobel(const oclMat &src, oclMat &dst, int ddepth, int dx, int dy, int ksize = 3, double scale = 1, double delta = 0.0, int bordertype = BORDER_DEFAULT);
802 //! applies the vertical or horizontal Scharr operator to the image
803 // dst.type must equalize src.type
804 // supports data type: CV_8UC1, CV_8UC4, CV_32FC1 and CV_32FC4
805 // supports border type: BORDER_CONSTANT, BORDER_REPLICATE, BORDER_REFLECT,BORDER_REFLECT_101
806 CV_EXPORTS void Scharr(const oclMat &src, oclMat &dst, int ddepth, int dx, int dy, double scale = 1, double delta = 0.0, int bordertype = BORDER_DEFAULT);
808 //! smooths the image using Gaussian filter.
809 // dst.type must equalize src.type
810 // supports data type: CV_8UC1, CV_8UC4, CV_32FC1 and CV_32FC4
811 // supports border type: BORDER_CONSTANT, BORDER_REPLICATE, BORDER_REFLECT,BORDER_REFLECT_101
812 CV_EXPORTS void GaussianBlur(const oclMat &src, oclMat &dst, Size ksize, double sigma1, double sigma2 = 0, int bordertype = BORDER_DEFAULT);
814 //! erodes the image (applies the local minimum operator)
815 // supports data type: CV_8UC1, CV_8UC4, CV_32FC1 and CV_32FC4
816 CV_EXPORTS void erode( const oclMat &src, oclMat &dst, const Mat &kernel, Point anchor = Point(-1, -1), int iterations = 1,
818 int borderType = BORDER_CONSTANT, const Scalar &borderValue = morphologyDefaultBorderValue());
821 //! dilates the image (applies the local maximum operator)
822 // supports data type: CV_8UC1, CV_8UC4, CV_32FC1 and CV_32FC4
823 CV_EXPORTS void dilate( const oclMat &src, oclMat &dst, const Mat &kernel, Point anchor = Point(-1, -1), int iterations = 1,
825 int borderType = BORDER_CONSTANT, const Scalar &borderValue = morphologyDefaultBorderValue());
828 //! applies an advanced morphological operation to the image
829 CV_EXPORTS void morphologyEx( const oclMat &src, oclMat &dst, int op, const Mat &kernel, Point anchor = Point(-1, -1), int iterations = 1,
831 int borderType = BORDER_CONSTANT, const Scalar &borderValue = morphologyDefaultBorderValue());
834 ////////////////////////////// Image processing //////////////////////////////
835 //! Does mean shift filtering on GPU.
836 CV_EXPORTS void meanShiftFiltering(const oclMat &src, oclMat &dst, int sp, int sr,
837 TermCriteria criteria = TermCriteria(TermCriteria::MAX_ITER + TermCriteria::EPS, 5, 1));
839 //! Does mean shift procedure on GPU.
840 CV_EXPORTS void meanShiftProc(const oclMat &src, oclMat &dstr, oclMat &dstsp, int sp, int sr,
841 TermCriteria criteria = TermCriteria(TermCriteria::MAX_ITER + TermCriteria::EPS, 5, 1));
843 //! Does mean shift segmentation with elimiation of small regions.
844 CV_EXPORTS void meanShiftSegmentation(const oclMat &src, Mat &dst, int sp, int sr, int minsize,
845 TermCriteria criteria = TermCriteria(TermCriteria::MAX_ITER + TermCriteria::EPS, 5, 1));
847 //! applies fixed threshold to the image.
848 // supports CV_8UC1 and CV_32FC1 data type
849 // supports threshold type: THRESH_BINARY, THRESH_BINARY_INV, THRESH_TRUNC, THRESH_TOZERO, THRESH_TOZERO_INV
850 CV_EXPORTS double threshold(const oclMat &src, oclMat &dst, double thresh, double maxVal, int type = THRESH_TRUNC);
852 //! resizes the image
853 // Supports INTER_NEAREST, INTER_LINEAR
854 // supports CV_8UC1, CV_8UC4, CV_32FC1 and CV_32FC4 types
855 CV_EXPORTS void resize(const oclMat &src, oclMat &dst, Size dsize, double fx = 0, double fy = 0, int interpolation = INTER_LINEAR);
857 //! Applies a generic geometrical transformation to an image.
859 // Supports INTER_NEAREST, INTER_LINEAR.
860 // Map1 supports CV_16SC2, CV_32FC2 types.
861 // Src supports CV_8UC1, CV_8UC2, CV_8UC4.
862 CV_EXPORTS void remap(const oclMat &src, oclMat &dst, oclMat &map1, oclMat &map2, int interpolation, int bordertype, const Scalar &value = Scalar());
864 //! copies 2D array to a larger destination array and pads borders with user-specifiable constant
865 // supports CV_8UC1, CV_8UC4, CV_32SC1 types
866 CV_EXPORTS void copyMakeBorder(const oclMat &src, oclMat &dst, int top, int bottom, int left, int right, int boardtype, const Scalar &value = Scalar());
868 //! Smoothes image using median filter
869 // The source 1- or 4-channel image. m should be 3 or 5, the image depth should be CV_8U or CV_32F.
870 CV_EXPORTS void medianFilter(const oclMat &src, oclMat &dst, int m);
872 //! warps the image using affine transformation
873 // Supports INTER_NEAREST, INTER_LINEAR, INTER_CUBIC
874 // supports CV_8UC1, CV_8UC4, CV_32FC1 and CV_32FC4 types
875 CV_EXPORTS void warpAffine(const oclMat &src, oclMat &dst, const Mat &M, Size dsize, int flags = INTER_LINEAR);
877 //! warps the image using perspective transformation
878 // Supports INTER_NEAREST, INTER_LINEAR, INTER_CUBIC
879 // supports CV_8UC1, CV_8UC4, CV_32FC1 and CV_32FC4 types
880 CV_EXPORTS void warpPerspective(const oclMat &src, oclMat &dst, const Mat &M, Size dsize, int flags = INTER_LINEAR);
882 //! computes the integral image and integral for the squared image
883 // sum will support CV_32S, CV_32F, sqsum - support CV32F, CV_64F
884 // supports only CV_8UC1 source type
885 CV_EXPORTS void integral(const oclMat &src, oclMat &sum, oclMat &sqsum, int sdepth=-1 );
886 CV_EXPORTS void integral(const oclMat &src, oclMat &sum, int sdepth=-1 );
887 CV_EXPORTS void cornerHarris(const oclMat &src, oclMat &dst, int blockSize, int ksize, double k, int bordertype = cv::BORDER_DEFAULT);
888 CV_EXPORTS void cornerHarris_dxdy(const oclMat &src, oclMat &dst, oclMat &Dx, oclMat &Dy,
889 int blockSize, int ksize, double k, int bordertype = cv::BORDER_DEFAULT);
890 CV_EXPORTS void cornerMinEigenVal(const oclMat &src, oclMat &dst, int blockSize, int ksize, int bordertype = cv::BORDER_DEFAULT);
891 CV_EXPORTS void cornerMinEigenVal_dxdy(const oclMat &src, oclMat &dst, oclMat &Dx, oclMat &Dy,
892 int blockSize, int ksize, int bordertype = cv::BORDER_DEFAULT);
895 /////////////////////////////////// ML ///////////////////////////////////////////
897 //! Compute closest centers for each lines in source and lable it after center's index
898 // supports CV_32FC1/CV_32FC2/CV_32FC4 data type
899 // supports NORM_L1 and NORM_L2 distType
900 // if indices is provided, only the indexed rows will be calculated and their results are in the same
902 CV_EXPORTS void distanceToCenters(const oclMat &src, const oclMat ¢ers, Mat &dists, Mat &labels, int distType = NORM_L2SQR);
904 //!Does k-means procedure on GPU
905 // supports CV_32FC1/CV_32FC2/CV_32FC4 data type
906 CV_EXPORTS double kmeans(const oclMat &src, int K, oclMat &bestLabels,
907 TermCriteria criteria, int attemps, int flags, oclMat ¢ers);
910 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
911 ///////////////////////////////////////////CascadeClassifier//////////////////////////////////////////////////////////////////
912 ///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
913 class CV_EXPORTS OclCascadeClassifier : public cv::CascadeClassifier
916 void detectMultiScale(oclMat &image, CV_OUT std::vector<cv::Rect>& faces,
917 double scaleFactor = 1.1, int minNeighbors = 3, int flags = 0,
918 Size minSize = Size(), Size maxSize = Size());
921 /////////////////////////////// Pyramid /////////////////////////////////////
922 CV_EXPORTS void pyrDown(const oclMat &src, oclMat &dst);
924 //! upsamples the source image and then smoothes it
925 CV_EXPORTS void pyrUp(const oclMat &src, oclMat &dst);
927 //! performs linear blending of two images
928 //! to avoid accuracy errors sum of weigths shouldn't be very close to zero
929 // supports only CV_8UC1 source type
930 CV_EXPORTS void blendLinear(const oclMat &img1, const oclMat &img2, const oclMat &weights1, const oclMat &weights2, oclMat &result);
932 //! computes vertical sum, supports only CV_32FC1 images
933 CV_EXPORTS void columnSum(const oclMat &src, oclMat &sum);
935 ///////////////////////////////////////// match_template /////////////////////////////////////////////////////////////
936 struct CV_EXPORTS MatchTemplateBuf
938 Size user_block_size;
939 oclMat imagef, templf;
940 std::vector<oclMat> images;
941 std::vector<oclMat> image_sums;
942 std::vector<oclMat> image_sqsums;
945 //! computes the proximity map for the raster template and the image where the template is searched for
946 // Supports TM_SQDIFF, TM_SQDIFF_NORMED, TM_CCORR, TM_CCORR_NORMED, TM_CCOEFF, TM_CCOEFF_NORMED for type 8UC1 and 8UC4
947 // Supports TM_SQDIFF, TM_CCORR for type 32FC1 and 32FC4
948 CV_EXPORTS void matchTemplate(const oclMat &image, const oclMat &templ, oclMat &result, int method);
950 //! computes the proximity map for the raster template and the image where the template is searched for
951 // Supports TM_SQDIFF, TM_SQDIFF_NORMED, TM_CCORR, TM_CCORR_NORMED, TM_CCOEFF, TM_CCOEFF_NORMED for type 8UC1 and 8UC4
952 // Supports TM_SQDIFF, TM_CCORR for type 32FC1 and 32FC4
953 CV_EXPORTS void matchTemplate(const oclMat &image, const oclMat &templ, oclMat &result, int method, MatchTemplateBuf &buf);
957 ///////////////////////////////////////////// Canny /////////////////////////////////////////////
958 struct CV_EXPORTS CannyBuf;
960 //! compute edges of the input image using Canny operator
961 // Support CV_8UC1 only
962 CV_EXPORTS void Canny(const oclMat &image, oclMat &edges, double low_thresh, double high_thresh, int apperture_size = 3, bool L2gradient = false);
963 CV_EXPORTS void Canny(const oclMat &image, CannyBuf &buf, oclMat &edges, double low_thresh, double high_thresh, int apperture_size = 3, bool L2gradient = false);
964 CV_EXPORTS void Canny(const oclMat &dx, const oclMat &dy, oclMat &edges, double low_thresh, double high_thresh, bool L2gradient = false);
965 CV_EXPORTS void Canny(const oclMat &dx, const oclMat &dy, CannyBuf &buf, oclMat &edges, double low_thresh, double high_thresh, bool L2gradient = false);
967 struct CV_EXPORTS CannyBuf
969 CannyBuf() : counter(1, 1, CV_32S) { }
974 explicit CannyBuf(const Size &image_size, int apperture_size = 3) : counter(1, 1, CV_32S)
976 create(image_size, apperture_size);
978 CannyBuf(const oclMat &dx_, const oclMat &dy_);
979 void create(const Size &image_size, int apperture_size = 3);
983 oclMat dx_buf, dy_buf;
984 oclMat magBuf, mapBuf;
985 oclMat trackBuf1, trackBuf2;
987 Ptr<FilterEngine_GPU> filterDX, filterDY;
990 ///////////////////////////////////////// Hough Transform /////////////////////////////////////////
992 struct HoughCirclesBuf
1001 CV_EXPORTS void HoughCircles(const oclMat& src, oclMat& circles, int method, float dp, float minDist, int cannyThreshold, int votesThreshold, int minRadius, int maxRadius, int maxCircles = 4096);
1002 CV_EXPORTS void HoughCircles(const oclMat& src, oclMat& circles, HoughCirclesBuf& buf, int method, float dp, float minDist, int cannyThreshold, int votesThreshold, int minRadius, int maxRadius, int maxCircles = 4096);
1003 CV_EXPORTS void HoughCirclesDownload(const oclMat& d_circles, OutputArray h_circles);
1006 ///////////////////////////////////////// clAmdFft related /////////////////////////////////////////
1007 //! Performs a forward or inverse discrete Fourier transform (1D or 2D) of floating point matrix.
1008 //! Param dft_size is the size of DFT transform.
1010 //! For complex-to-real transform it is assumed that the source matrix is packed in CLFFT's format.
1011 // support src type of CV32FC1, CV32FC2
1012 // support flags: DFT_INVERSE, DFT_REAL_OUTPUT, DFT_COMPLEX_OUTPUT, DFT_ROWS
1013 // dft_size is the size of original input, which is used for transformation from complex to real.
1014 // dft_size must be powers of 2, 3 and 5
1015 // real to complex dft requires at least v1.8 clAmdFft
1016 // real to complex dft output is not the same with cpu version
1017 // real to complex and complex to real does not support DFT_ROWS
1018 CV_EXPORTS void dft(const oclMat &src, oclMat &dst, Size dft_size = Size(), int flags = 0);
1020 //! implements generalized matrix product algorithm GEMM from BLAS
1021 // The functionality requires clAmdBlas library
1022 // only support type CV_32FC1
1023 // flag GEMM_3_T is not supported
1024 CV_EXPORTS void gemm(const oclMat &src1, const oclMat &src2, double alpha,
1025 const oclMat &src3, double beta, oclMat &dst, int flags = 0);
1027 //////////////// HOG (Histogram-of-Oriented-Gradients) Descriptor and Object Detector //////////////
1029 struct CV_EXPORTS HOGDescriptor
1033 enum { DEFAULT_WIN_SIGMA = -1 };
1035 enum { DEFAULT_NLEVELS = 64 };
1037 enum { DESCR_FORMAT_ROW_BY_ROW, DESCR_FORMAT_COL_BY_COL };
1041 HOGDescriptor(Size win_size = Size(64, 128), Size block_size = Size(16, 16),
1043 Size block_stride = Size(8, 8), Size cell_size = Size(8, 8),
1045 int nbins = 9, double win_sigma = DEFAULT_WIN_SIGMA,
1047 double threshold_L2hys = 0.2, bool gamma_correction = true,
1049 int nlevels = DEFAULT_NLEVELS);
1053 size_t getDescriptorSize() const;
1055 size_t getBlockHistogramSize() const;
1059 void setSVMDetector(const std::vector<float> &detector);
1063 static std::vector<float> getDefaultPeopleDetector();
1065 static std::vector<float> getPeopleDetector48x96();
1067 static std::vector<float> getPeopleDetector64x128();
1071 void detect(const oclMat &img, std::vector<Point> &found_locations,
1073 double hit_threshold = 0, Size win_stride = Size(),
1075 Size padding = Size());
1079 void detectMultiScale(const oclMat &img, std::vector<Rect> &found_locations,
1081 double hit_threshold = 0, Size win_stride = Size(),
1083 Size padding = Size(), double scale0 = 1.05,
1085 int group_threshold = 2);
1089 void getDescriptors(const oclMat &img, Size win_stride,
1091 oclMat &descriptors,
1093 int descr_format = DESCR_FORMAT_COL_BY_COL);
1109 double threshold_L2hys;
1111 bool gamma_correction;
1119 // initialize buffers; only need to do once in case of multiscale detection
1121 void init_buffer(const oclMat &img, Size win_stride);
1125 void computeBlockHistograms(const oclMat &img);
1127 void computeGradient(const oclMat &img, oclMat &grad, oclMat &qangle);
1131 double getWinSigma() const;
1133 bool checkDetectorSize() const;
1137 static int numPartsWithin(int size, int part_size, int stride);
1139 static Size numPartsWithin(Size size, Size part_size, Size stride);
1143 // Coefficients of the separating plane
1151 // Results of the last classification step
1159 // Results of the last histogram evaluation step
1165 // Gradients conputation results
1167 oclMat grad, qangle;
1177 // effect size of input image (might be different from original size after scaling)
1184 ////////////////////////feature2d_ocl/////////////////
1185 /****************************************************************************************\
1187 \****************************************************************************************/
1188 template<typename T>
1189 struct CV_EXPORTS Accumulator
1193 template<> struct Accumulator<unsigned char>
1197 template<> struct Accumulator<unsigned short>
1201 template<> struct Accumulator<char>
1205 template<> struct Accumulator<short>
1211 * Manhattan distance (city block distance) functor
1214 struct CV_EXPORTS L1
1216 enum { normType = NORM_L1 };
1217 typedef T ValueType;
1218 typedef typename Accumulator<T>::Type ResultType;
1220 ResultType operator()( const T *a, const T *b, int size ) const
1222 return normL1<ValueType, ResultType>(a, b, size);
1227 * Euclidean distance functor
1230 struct CV_EXPORTS L2
1232 enum { normType = NORM_L2 };
1233 typedef T ValueType;
1234 typedef typename Accumulator<T>::Type ResultType;
1236 ResultType operator()( const T *a, const T *b, int size ) const
1238 return (ResultType)std::sqrt((double)normL2Sqr<ValueType, ResultType>(a, b, size));
1243 * Hamming distance functor - counts the bit differences between two strings - useful for the Brief descriptor
1244 * bit count of A exclusive XOR'ed with B
1246 struct CV_EXPORTS Hamming
1248 enum { normType = NORM_HAMMING };
1249 typedef unsigned char ValueType;
1250 typedef int ResultType;
1252 /** this will count the bits in a ^ b
1254 ResultType operator()( const unsigned char *a, const unsigned char *b, int size ) const
1256 return normHamming(a, b, size);
1260 ////////////////////////////////// BruteForceMatcher //////////////////////////////////
1262 class CV_EXPORTS BruteForceMatcher_OCL_base
1265 enum DistType {L1Dist = 0, L2Dist, HammingDist};
1266 explicit BruteForceMatcher_OCL_base(DistType distType = L2Dist);
1268 // Add descriptors to train descriptor collection
1269 void add(const std::vector<oclMat> &descCollection);
1271 // Get train descriptors collection
1272 const std::vector<oclMat> &getTrainDescriptors() const;
1274 // Clear train descriptors collection
1277 // Return true if there are not train descriptors in collection
1280 // Return true if the matcher supports mask in match methods
1281 bool isMaskSupported() const;
1283 // Find one best match for each query descriptor
1284 void matchSingle(const oclMat &query, const oclMat &train,
1285 oclMat &trainIdx, oclMat &distance,
1286 const oclMat &mask = oclMat());
1288 // Download trainIdx and distance and convert it to CPU vector with DMatch
1289 static void matchDownload(const oclMat &trainIdx, const oclMat &distance, std::vector<DMatch> &matches);
1290 // Convert trainIdx and distance to vector with DMatch
1291 static void matchConvert(const Mat &trainIdx, const Mat &distance, std::vector<DMatch> &matches);
1293 // Find one best match for each query descriptor
1294 void match(const oclMat &query, const oclMat &train, std::vector<DMatch> &matches, const oclMat &mask = oclMat());
1296 // Make gpu collection of trains and masks in suitable format for matchCollection function
1297 void makeGpuCollection(oclMat &trainCollection, oclMat &maskCollection, const std::vector<oclMat> &masks = std::vector<oclMat>());
1299 // Find one best match from train collection for each query descriptor
1300 void matchCollection(const oclMat &query, const oclMat &trainCollection,
1301 oclMat &trainIdx, oclMat &imgIdx, oclMat &distance,
1302 const oclMat &masks = oclMat());
1304 // Download trainIdx, imgIdx and distance and convert it to vector with DMatch
1305 static void matchDownload(const oclMat &trainIdx, const oclMat &imgIdx, const oclMat &distance, std::vector<DMatch> &matches);
1306 // Convert trainIdx, imgIdx and distance to vector with DMatch
1307 static void matchConvert(const Mat &trainIdx, const Mat &imgIdx, const Mat &distance, std::vector<DMatch> &matches);
1309 // Find one best match from train collection for each query descriptor.
1310 void match(const oclMat &query, std::vector<DMatch> &matches, const std::vector<oclMat> &masks = std::vector<oclMat>());
1312 // Find k best matches for each query descriptor (in increasing order of distances)
1313 void knnMatchSingle(const oclMat &query, const oclMat &train,
1314 oclMat &trainIdx, oclMat &distance, oclMat &allDist, int k,
1315 const oclMat &mask = oclMat());
1317 // Download trainIdx and distance and convert it to vector with DMatch
1318 // compactResult is used when mask is not empty. If compactResult is false matches
1319 // vector will have the same size as queryDescriptors rows. If compactResult is true
1320 // matches vector will not contain matches for fully masked out query descriptors.
1321 static void knnMatchDownload(const oclMat &trainIdx, const oclMat &distance,
1322 std::vector< std::vector<DMatch> > &matches, bool compactResult = false);
1323 // Convert trainIdx and distance to vector with DMatch
1324 static void knnMatchConvert(const Mat &trainIdx, const Mat &distance,
1325 std::vector< std::vector<DMatch> > &matches, bool compactResult = false);
1327 // Find k best matches for each query descriptor (in increasing order of distances).
1328 // compactResult is used when mask is not empty. If compactResult is false matches
1329 // vector will have the same size as queryDescriptors rows. If compactResult is true
1330 // matches vector will not contain matches for fully masked out query descriptors.
1331 void knnMatch(const oclMat &query, const oclMat &train,
1332 std::vector< std::vector<DMatch> > &matches, int k, const oclMat &mask = oclMat(),
1333 bool compactResult = false);
1335 // Find k best matches from train collection for each query descriptor (in increasing order of distances)
1336 void knnMatch2Collection(const oclMat &query, const oclMat &trainCollection,
1337 oclMat &trainIdx, oclMat &imgIdx, oclMat &distance,
1338 const oclMat &maskCollection = oclMat());
1340 // Download trainIdx and distance and convert it to vector with DMatch
1341 // compactResult is used when mask is not empty. If compactResult is false matches
1342 // vector will have the same size as queryDescriptors rows. If compactResult is true
1343 // matches vector will not contain matches for fully masked out query descriptors.
1344 static void knnMatch2Download(const oclMat &trainIdx, const oclMat &imgIdx, const oclMat &distance,
1345 std::vector< std::vector<DMatch> > &matches, bool compactResult = false);
1346 // Convert trainIdx and distance to vector with DMatch
1347 static void knnMatch2Convert(const Mat &trainIdx, const Mat &imgIdx, const Mat &distance,
1348 std::vector< std::vector<DMatch> > &matches, bool compactResult = false);
1350 // Find k best matches for each query descriptor (in increasing order of distances).
1351 // compactResult is used when mask is not empty. If compactResult is false matches
1352 // vector will have the same size as queryDescriptors rows. If compactResult is true
1353 // matches vector will not contain matches for fully masked out query descriptors.
1354 void knnMatch(const oclMat &query, std::vector< std::vector<DMatch> > &matches, int k,
1355 const std::vector<oclMat> &masks = std::vector<oclMat>(), bool compactResult = false);
1357 // Find best matches for each query descriptor which have distance less than maxDistance.
1358 // nMatches.at<int>(0, queryIdx) will contain matches count for queryIdx.
1359 // carefully nMatches can be greater than trainIdx.cols - it means that matcher didn't find all matches,
1360 // because it didn't have enough memory.
1361 // If trainIdx is empty, then trainIdx and distance will be created with size nQuery x max((nTrain / 100), 10),
1362 // otherwize user can pass own allocated trainIdx and distance with size nQuery x nMaxMatches
1363 // Matches doesn't sorted.
1364 void radiusMatchSingle(const oclMat &query, const oclMat &train,
1365 oclMat &trainIdx, oclMat &distance, oclMat &nMatches, float maxDistance,
1366 const oclMat &mask = oclMat());
1368 // Download trainIdx, nMatches and distance and convert it to vector with DMatch.
1369 // matches will be sorted in increasing order of distances.
1370 // compactResult is used when mask is not empty. If compactResult is false matches
1371 // vector will have the same size as queryDescriptors rows. If compactResult is true
1372 // matches vector will not contain matches for fully masked out query descriptors.
1373 static void radiusMatchDownload(const oclMat &trainIdx, const oclMat &distance, const oclMat &nMatches,
1374 std::vector< std::vector<DMatch> > &matches, bool compactResult = false);
1375 // Convert trainIdx, nMatches and distance to vector with DMatch.
1376 static void radiusMatchConvert(const Mat &trainIdx, const Mat &distance, const Mat &nMatches,
1377 std::vector< std::vector<DMatch> > &matches, bool compactResult = false);
1379 // Find best matches for each query descriptor which have distance less than maxDistance
1380 // in increasing order of distances).
1381 void radiusMatch(const oclMat &query, const oclMat &train,
1382 std::vector< std::vector<DMatch> > &matches, float maxDistance,
1383 const oclMat &mask = oclMat(), bool compactResult = false);
1385 // Find best matches for each query descriptor which have distance less than maxDistance.
1386 // If trainIdx is empty, then trainIdx and distance will be created with size nQuery x max((nQuery / 100), 10),
1387 // otherwize user can pass own allocated trainIdx and distance with size nQuery x nMaxMatches
1388 // Matches doesn't sorted.
1389 void radiusMatchCollection(const oclMat &query, oclMat &trainIdx, oclMat &imgIdx, oclMat &distance, oclMat &nMatches, float maxDistance,
1390 const std::vector<oclMat> &masks = std::vector<oclMat>());
1392 // Download trainIdx, imgIdx, nMatches and distance and convert it to vector with DMatch.
1393 // matches will be sorted in increasing order of distances.
1394 // compactResult is used when mask is not empty. If compactResult is false matches
1395 // vector will have the same size as queryDescriptors rows. If compactResult is true
1396 // matches vector will not contain matches for fully masked out query descriptors.
1397 static void radiusMatchDownload(const oclMat &trainIdx, const oclMat &imgIdx, const oclMat &distance, const oclMat &nMatches,
1398 std::vector< std::vector<DMatch> > &matches, bool compactResult = false);
1399 // Convert trainIdx, nMatches and distance to vector with DMatch.
1400 static void radiusMatchConvert(const Mat &trainIdx, const Mat &imgIdx, const Mat &distance, const Mat &nMatches,
1401 std::vector< std::vector<DMatch> > &matches, bool compactResult = false);
1403 // Find best matches from train collection for each query descriptor which have distance less than
1404 // maxDistance (in increasing order of distances).
1405 void radiusMatch(const oclMat &query, std::vector< std::vector<DMatch> > &matches, float maxDistance,
1406 const std::vector<oclMat> &masks = std::vector<oclMat>(), bool compactResult = false);
1411 std::vector<oclMat> trainDescCollection;
1414 template <class Distance>
1415 class CV_EXPORTS BruteForceMatcher_OCL;
1417 template <typename T>
1418 class CV_EXPORTS BruteForceMatcher_OCL< L1<T> > : public BruteForceMatcher_OCL_base
1421 explicit BruteForceMatcher_OCL() : BruteForceMatcher_OCL_base(L1Dist) {}
1422 explicit BruteForceMatcher_OCL(L1<T> /*d*/) : BruteForceMatcher_OCL_base(L1Dist) {}
1424 template <typename T>
1425 class CV_EXPORTS BruteForceMatcher_OCL< L2<T> > : public BruteForceMatcher_OCL_base
1428 explicit BruteForceMatcher_OCL() : BruteForceMatcher_OCL_base(L2Dist) {}
1429 explicit BruteForceMatcher_OCL(L2<T> /*d*/) : BruteForceMatcher_OCL_base(L2Dist) {}
1431 template <> class CV_EXPORTS BruteForceMatcher_OCL< Hamming > : public BruteForceMatcher_OCL_base
1434 explicit BruteForceMatcher_OCL() : BruteForceMatcher_OCL_base(HammingDist) {}
1435 explicit BruteForceMatcher_OCL(Hamming /*d*/) : BruteForceMatcher_OCL_base(HammingDist) {}
1438 class CV_EXPORTS BFMatcher_OCL : public BruteForceMatcher_OCL_base
1441 explicit BFMatcher_OCL(int norm = NORM_L2) : BruteForceMatcher_OCL_base(norm == NORM_L1 ? L1Dist : norm == NORM_L2 ? L2Dist : HammingDist) {}
1444 class CV_EXPORTS GoodFeaturesToTrackDetector_OCL
1447 explicit GoodFeaturesToTrackDetector_OCL(int maxCorners = 1000, double qualityLevel = 0.01, double minDistance = 0.0,
1448 int blockSize = 3, bool useHarrisDetector = false, double harrisK = 0.04);
1450 //! return 1 rows matrix with CV_32FC2 type
1451 void operator ()(const oclMat& image, oclMat& corners, const oclMat& mask = oclMat());
1452 //! download points of type Point2f to a vector. the vector's content will be erased
1453 void downloadPoints(const oclMat &points, std::vector<Point2f> &points_v);
1456 double qualityLevel;
1460 bool useHarrisDetector;
1462 void releaseMemory()
1467 minMaxbuf_.release();
1468 tmpCorners_.release();
1478 inline GoodFeaturesToTrackDetector_OCL::GoodFeaturesToTrackDetector_OCL(int maxCorners_, double qualityLevel_, double minDistance_,
1479 int blockSize_, bool useHarrisDetector_, double harrisK_)
1481 maxCorners = maxCorners_;
1482 qualityLevel = qualityLevel_;
1483 minDistance = minDistance_;
1484 blockSize = blockSize_;
1485 useHarrisDetector = useHarrisDetector_;
1489 ////////////////////////////////// FAST Feature Detector //////////////////////////////////
1490 class CV_EXPORTS FAST_OCL
1501 // all features have same size
1502 static const int FEATURE_SIZE = 7;
1504 explicit FAST_OCL(int threshold, bool nonmaxSupression = true, double keypointsRatio = 0.05);
1506 //! finds the keypoints using FAST detector
1507 //! supports only CV_8UC1 images
1508 void operator ()(const oclMat& image, const oclMat& mask, oclMat& keypoints);
1509 void operator ()(const oclMat& image, const oclMat& mask, std::vector<KeyPoint>& keypoints);
1511 //! download keypoints from device to host memory
1512 static void downloadKeypoints(const oclMat& d_keypoints, std::vector<KeyPoint>& keypoints);
1514 //! convert keypoints to KeyPoint vector
1515 static void convertKeypoints(const Mat& h_keypoints, std::vector<KeyPoint>& keypoints);
1517 //! release temporary buffer's memory
1520 bool nonmaxSupression;
1524 //! max keypoints = keypointsRatio * img.size().area()
1525 double keypointsRatio;
1527 //! find keypoints and compute it's response if nonmaxSupression is true
1528 //! return count of detected keypoints
1529 int calcKeyPointsLocation(const oclMat& image, const oclMat& mask);
1531 //! get final array of keypoints
1532 //! performs nonmax supression if needed
1533 //! return final count of keypoints
1534 int getKeyPoints(oclMat& keypoints);
1542 oclMat d_keypoints_;
1544 int calcKeypointsOCL(const oclMat& img, const oclMat& mask, int maxKeypoints);
1545 int nonmaxSupressionOCL(oclMat& keypoints);
1548 /////////////////////////////// PyrLKOpticalFlow /////////////////////////////////////
1550 class CV_EXPORTS PyrLKOpticalFlow
1555 winSize = Size(21, 21);
1559 useInitialFlow = false;
1560 minEigThreshold = 1e-4f;
1561 getMinEigenVals = false;
1562 isDeviceArch11_ = false;
1565 void sparse(const oclMat &prevImg, const oclMat &nextImg, const oclMat &prevPts, oclMat &nextPts,
1566 oclMat &status, oclMat *err = 0);
1568 void dense(const oclMat &prevImg, const oclMat &nextImg, oclMat &u, oclMat &v, oclMat *err = 0);
1574 bool useInitialFlow;
1575 float minEigThreshold;
1576 bool getMinEigenVals;
1578 void releaseMemory()
1580 dx_calcBuf_.release();
1581 dy_calcBuf_.release();
1591 void calcSharrDeriv(const oclMat &src, oclMat &dx, oclMat &dy);
1593 void buildImagePyramid(const oclMat &img0, std::vector<oclMat> &pyr, bool withBorder);
1598 std::vector<oclMat> prevPyr_;
1599 std::vector<oclMat> nextPyr_;
1607 bool isDeviceArch11_;
1610 class CV_EXPORTS FarnebackOpticalFlow
1613 FarnebackOpticalFlow();
1624 void operator ()(const oclMat &frame0, const oclMat &frame1, oclMat &flowx, oclMat &flowy);
1626 void releaseMemory();
1629 void prepareGaussian(
1630 int n, double sigma, float *g, float *xg, float *xxg,
1631 double &ig11, double &ig03, double &ig33, double &ig55);
1633 void setPolynomialExpansionConsts(int n, double sigma);
1635 void updateFlow_boxFilter(
1636 const oclMat& R0, const oclMat& R1, oclMat& flowx, oclMat &flowy,
1637 oclMat& M, oclMat &bufM, int blockSize, bool updateMatrices);
1639 void updateFlow_gaussianBlur(
1640 const oclMat& R0, const oclMat& R1, oclMat& flowx, oclMat& flowy,
1641 oclMat& M, oclMat &bufM, int blockSize, bool updateMatrices);
1644 oclMat pyrLevel_[2], M_, bufM_, R_[2], blurredFrame_[2];
1645 std::vector<oclMat> pyramid0_, pyramid1_;
1648 //////////////// build warping maps ////////////////////
1649 //! builds plane warping maps
1650 CV_EXPORTS void buildWarpPlaneMaps(Size src_size, Rect dst_roi, const Mat &K, const Mat &R, const Mat &T, float scale, oclMat &map_x, oclMat &map_y);
1651 //! builds cylindrical warping maps
1652 CV_EXPORTS void buildWarpCylindricalMaps(Size src_size, Rect dst_roi, const Mat &K, const Mat &R, float scale, oclMat &map_x, oclMat &map_y);
1653 //! builds spherical warping maps
1654 CV_EXPORTS void buildWarpSphericalMaps(Size src_size, Rect dst_roi, const Mat &K, const Mat &R, float scale, oclMat &map_x, oclMat &map_y);
1655 //! builds Affine warping maps
1656 CV_EXPORTS void buildWarpAffineMaps(const Mat &M, bool inverse, Size dsize, oclMat &xmap, oclMat &ymap);
1658 //! builds Perspective warping maps
1659 CV_EXPORTS void buildWarpPerspectiveMaps(const Mat &M, bool inverse, Size dsize, oclMat &xmap, oclMat &ymap);
1661 ///////////////////////////////////// interpolate frames //////////////////////////////////////////////
1662 //! Interpolate frames (images) using provided optical flow (displacement field).
1663 //! frame0 - frame 0 (32-bit floating point images, single channel)
1664 //! frame1 - frame 1 (the same type and size)
1665 //! fu - forward horizontal displacement
1666 //! fv - forward vertical displacement
1667 //! bu - backward horizontal displacement
1668 //! bv - backward vertical displacement
1669 //! pos - new frame position
1670 //! newFrame - new frame
1671 //! buf - temporary buffer, will have width x 6*height size, CV_32FC1 type and contain 6 oclMat;
1672 //! occlusion masks 0, occlusion masks 1,
1673 //! interpolated forward flow 0, interpolated forward flow 1,
1674 //! interpolated backward flow 0, interpolated backward flow 1
1676 CV_EXPORTS void interpolateFrames(const oclMat &frame0, const oclMat &frame1,
1677 const oclMat &fu, const oclMat &fv,
1678 const oclMat &bu, const oclMat &bv,
1679 float pos, oclMat &newFrame, oclMat &buf);
1681 //! computes moments of the rasterized shape or a vector of points
1682 //! _array should be a vector a points standing for the contour
1683 CV_EXPORTS Moments ocl_moments(InputArray contour);
1684 //! src should be a general image uploaded to the GPU.
1685 //! the supported oclMat type are CV_8UC1, CV_16UC1, CV_16SC1, CV_32FC1 and CV_64FC1
1686 //! to use type of CV_64FC1, the GPU should support CV_64FC1
1687 CV_EXPORTS Moments ocl_moments(oclMat& src, bool binary);
1689 class CV_EXPORTS StereoBM_OCL
1692 enum { BASIC_PRESET = 0, PREFILTER_XSOBEL = 1 };
1694 enum { DEFAULT_NDISP = 64, DEFAULT_WINSZ = 19 };
1696 //! the default constructor
1698 //! the full constructor taking the camera-specific preset, number of disparities and the SAD window size. ndisparities must be multiple of 8.
1699 StereoBM_OCL(int preset, int ndisparities = DEFAULT_NDISP, int winSize = DEFAULT_WINSZ);
1701 //! the stereo correspondence operator. Finds the disparity for the specified rectified stereo pair
1702 //! Output disparity has CV_8U type.
1703 void operator() ( const oclMat &left, const oclMat &right, oclMat &disparity);
1705 //! Some heuristics that tries to estmate
1706 // if current GPU will be faster then CPU in this algorithm.
1707 // It queries current active device.
1708 static bool checkIfGpuCallReasonable();
1714 // If avergeTexThreshold == 0 => post procesing is disabled
1715 // If avergeTexThreshold != 0 then disparity is set 0 in each point (x,y) where for left image
1716 // SumOfHorizontalGradiensInWindow(x, y, winSize) < (winSize * winSize) * avergeTexThreshold
1717 // i.e. input left image is low textured.
1718 float avergeTexThreshold;
1720 oclMat minSSD, leBuf, riBuf;
1723 class CV_EXPORTS StereoBeliefPropagation
1726 enum { DEFAULT_NDISP = 64 };
1727 enum { DEFAULT_ITERS = 5 };
1728 enum { DEFAULT_LEVELS = 5 };
1729 static void estimateRecommendedParams(int width, int height, int &ndisp, int &iters, int &levels);
1730 explicit StereoBeliefPropagation(int ndisp = DEFAULT_NDISP,
1731 int iters = DEFAULT_ITERS,
1732 int levels = DEFAULT_LEVELS,
1733 int msg_type = CV_16S);
1734 StereoBeliefPropagation(int ndisp, int iters, int levels,
1735 float max_data_term, float data_weight,
1736 float max_disc_term, float disc_single_jump,
1737 int msg_type = CV_32F);
1738 void operator()(const oclMat &left, const oclMat &right, oclMat &disparity);
1739 void operator()(const oclMat &data, oclMat &disparity);
1743 float max_data_term;
1745 float max_disc_term;
1746 float disc_single_jump;
1749 oclMat u, d, l, r, u2, d2, l2, r2;
1750 std::vector<oclMat> datas;
1754 class CV_EXPORTS StereoConstantSpaceBP
1757 enum { DEFAULT_NDISP = 128 };
1758 enum { DEFAULT_ITERS = 8 };
1759 enum { DEFAULT_LEVELS = 4 };
1760 enum { DEFAULT_NR_PLANE = 4 };
1761 static void estimateRecommendedParams(int width, int height, int &ndisp, int &iters, int &levels, int &nr_plane);
1762 explicit StereoConstantSpaceBP(
1763 int ndisp = DEFAULT_NDISP,
1764 int iters = DEFAULT_ITERS,
1765 int levels = DEFAULT_LEVELS,
1766 int nr_plane = DEFAULT_NR_PLANE,
1767 int msg_type = CV_32F);
1768 StereoConstantSpaceBP(int ndisp, int iters, int levels, int nr_plane,
1769 float max_data_term, float data_weight, float max_disc_term, float disc_single_jump,
1770 int min_disp_th = 0,
1771 int msg_type = CV_32F);
1772 void operator()(const oclMat &left, const oclMat &right, oclMat &disparity);
1777 float max_data_term;
1779 float max_disc_term;
1780 float disc_single_jump;
1783 bool use_local_init_data_cost;
1785 oclMat u[2], d[2], l[2], r[2];
1786 oclMat disp_selected_pyr[2];
1788 oclMat data_cost_selected;
1793 // Implementation of the Zach, Pock and Bischof Dual TV-L1 Optical Flow method
1796 // [1] C. Zach, T. Pock and H. Bischof, "A Duality Based Approach for Realtime TV-L1 Optical Flow".
1797 // [2] Javier Sanchez, Enric Meinhardt-Llopis and Gabriele Facciolo. "TV-L1 Optical Flow Estimation".
1798 class CV_EXPORTS OpticalFlowDual_TVL1_OCL
1801 OpticalFlowDual_TVL1_OCL();
1803 void operator ()(const oclMat& I0, const oclMat& I1, oclMat& flowx, oclMat& flowy);
1805 void collectGarbage();
1808 * Time step of the numerical scheme.
1813 * Weight parameter for the data term, attachment parameter.
1814 * This is the most relevant parameter, which determines the smoothness of the output.
1815 * The smaller this parameter is, the smoother the solutions we obtain.
1816 * It depends on the range of motions of the images, so its value should be adapted to each image sequence.
1821 * Weight parameter for (u - v)^2, tightness parameter.
1822 * It serves as a link between the attachment and the regularization terms.
1823 * In theory, it should have a small value in order to maintain both parts in correspondence.
1824 * The method is stable for a large range of values of this parameter.
1829 * Number of scales used to create the pyramid of images.
1834 * Number of warpings per scale.
1835 * Represents the number of times that I1(x+u0) and grad( I1(x+u0) ) are computed per scale.
1836 * This is a parameter that assures the stability of the method.
1837 * It also affects the running time, so it is a compromise between speed and accuracy.
1842 * Stopping criterion threshold used in the numerical scheme, which is a trade-off between precision and running time.
1843 * A small value will yield more accurate solutions at the expense of a slower convergence.
1848 * Stopping criterion iterations number used in the numerical scheme.
1852 bool useInitialFlow;
1855 void procOneScale(const oclMat& I0, const oclMat& I1, oclMat& u1, oclMat& u2);
1857 std::vector<oclMat> I0s;
1858 std::vector<oclMat> I1s;
1859 std::vector<oclMat> u1s;
1860 std::vector<oclMat> u2s;
1880 // current supported sorting methods
1883 SORT_BITONIC, // only support power-of-2 buffer size
1884 SORT_SELECTION, // cannot sort duplicate keys
1886 SORT_RADIX // only support signed int/float keys(CV_32S/CV_32F)
1888 //! Returns the sorted result of all the elements in input based on equivalent keys.
1890 // The element unit in the values to be sorted is determined from the data type,
1891 // i.e., a CV_32FC2 input {a1a2, b1b2} will be considered as two elements, regardless its
1892 // matrix dimension.
1893 // both keys and values will be sorted inplace
1894 // Key needs to be single channel oclMat.
1898 // keys = {2, 3, 1} (CV_8UC1)
1899 // values = {10,5, 4,3, 6,2} (CV_8UC2)
1900 // sortByKey(keys, values, SORT_SELECTION, false);
1902 // keys = {1, 2, 3} (CV_8UC1)
1903 // values = {6,2, 10,5, 4,3} (CV_8UC2)
1904 CV_EXPORTS void sortByKey(oclMat& keys, oclMat& values, int method, bool isGreaterThan = false);
1905 /*!Base class for MOG and MOG2!*/
1906 class CV_EXPORTS BackgroundSubtractor
1909 //! the virtual destructor
1910 virtual ~BackgroundSubtractor();
1911 //! the update operator that takes the next video frame and returns the current foreground mask as 8-bit binary image.
1912 virtual void operator()(const oclMat& image, oclMat& fgmask, float learningRate);
1914 //! computes a background image
1915 virtual void getBackgroundImage(oclMat& backgroundImage) const = 0;
1918 Gaussian Mixture-based Backbround/Foreground Segmentation Algorithm
1920 The class implements the following algorithm:
1921 "An improved adaptive background mixture model for real-time tracking with shadow detection"
1922 P. KadewTraKuPong and R. Bowden,
1923 Proc. 2nd European Workshp on Advanced Video-Based Surveillance Systems, 2001."
1924 http://personal.ee.surrey.ac.uk/Personal/R.Bowden/publications/avbs01/avbs01.pdf
1926 class CV_EXPORTS MOG: public cv::ocl::BackgroundSubtractor
1929 //! the default constructor
1930 MOG(int nmixtures = -1);
1932 //! re-initiaization method
1933 void initialize(Size frameSize, int frameType);
1935 //! the update operator
1936 void operator()(const oclMat& frame, oclMat& fgmask, float learningRate = 0.f);
1938 //! computes a background image which are the mean of all background gaussians
1939 void getBackgroundImage(oclMat& backgroundImage) const;
1941 //! releases all inner buffers
1946 float backgroundRatio;
1963 The class implements the following algorithm:
1964 "Improved adaptive Gausian mixture model for background subtraction"
1966 International Conference Pattern Recognition, UK, August, 2004.
1967 http://www.zoranz.net/Publications/zivkovic2004ICPR.pdf
1969 class CV_EXPORTS MOG2: public cv::ocl::BackgroundSubtractor
1972 //! the default constructor
1973 MOG2(int nmixtures = -1);
1975 //! re-initiaization method
1976 void initialize(Size frameSize, int frameType);
1978 //! the update operator
1979 void operator()(const oclMat& frame, oclMat& fgmask, float learningRate = -1.0f);
1981 //! computes a background image which are the mean of all background gaussians
1982 void getBackgroundImage(oclMat& backgroundImage) const;
1984 //! releases all inner buffers
1988 // you should call initialize after parameters changes
1992 //! here it is the maximum allowed number of mixture components.
1993 //! Actual number is determined dynamically per pixel
1995 // threshold on the squared Mahalanobis distance to decide if it is well described
1996 // by the background model or not. Related to Cthr from the paper.
1997 // This does not influence the update of the background. A typical value could be 4 sigma
1998 // and that is varThreshold=4*4=16; Corresponds to Tb in the paper.
2000 /////////////////////////
2001 // less important parameters - things you might change but be carefull
2002 ////////////////////////
2004 float backgroundRatio;
2005 // corresponds to fTB=1-cf from the paper
2006 // TB - threshold when the component becomes significant enough to be included into
2007 // the background model. It is the TB=1-cf from the paper. So I use cf=0.1 => TB=0.
2008 // For alpha=0.001 it means that the mode should exist for approximately 105 frames before
2009 // it is considered foreground
2010 // float noiseSigma;
2011 float varThresholdGen;
2013 //correspondts to Tg - threshold on the squared Mahalan. dist. to decide
2014 //when a sample is close to the existing components. If it is not close
2015 //to any a new component will be generated. I use 3 sigma => Tg=3*3=9.
2016 //Smaller Tg leads to more generated components and higher Tg might make
2017 //lead to small number of components but they can grow too large
2022 //initial variance for the newly generated components.
2023 //It will will influence the speed of adaptation. A good guess should be made.
2024 //A simple way is to estimate the typical standard deviation from the images.
2025 //I used here 10 as a reasonable value
2026 // min and max can be used to further control the variance
2027 float fCT; //CT - complexity reduction prior
2028 //this is related to the number of samples needed to accept that a component
2029 //actually exists. We use CT=0.05 of all the samples. By setting CT=0 you get
2030 //the standard Stauffer&Grimson algorithm (maybe not exact but very similar)
2032 //shadow detection parameters
2033 bool bShadowDetection; //default 1 - do shadow detection
2034 unsigned char nShadowDetection; //do shadow detection - insert this value as the detection result - 127 default value
2036 // Tau - shadow threshold. The shadow is detected if the pixel is darker
2037 //version of the background. Tau is a threshold on how much darker the shadow can be.
2038 //Tau= 0.5 means that if pixel is more than 2 times darker then it is not shadow
2039 //See: Prati,Mikic,Trivedi,Cucchiarra,"Detecting Moving Shadows...",IEEE PAMI,2003.
2052 oclMat bgmodelUsedModes_; //keep track of number of modes per pixel
2055 /*!***************Kalman Filter*************!*/
2056 class CV_EXPORTS KalmanFilter
2060 //! the full constructor taking the dimensionality of the state, of the measurement and of the control vector
2061 KalmanFilter(int dynamParams, int measureParams, int controlParams=0, int type=CV_32F);
2062 //! re-initializes Kalman filter. The previous content is destroyed.
2063 void init(int dynamParams, int measureParams, int controlParams=0, int type=CV_32F);
2065 const oclMat& predict(const oclMat& control=oclMat());
2066 const oclMat& correct(const oclMat& measurement);
2068 oclMat statePre; //!< predicted state (x'(k)): x(k)=A*x(k-1)+B*u(k)
2069 oclMat statePost; //!< corrected state (x(k)): x(k)=x'(k)+K(k)*(z(k)-H*x'(k))
2070 oclMat transitionMatrix; //!< state transition matrix (A)
2071 oclMat controlMatrix; //!< control matrix (B) (not used if there is no control)
2072 oclMat measurementMatrix; //!< measurement matrix (H)
2073 oclMat processNoiseCov; //!< process noise covariance matrix (Q)
2074 oclMat measurementNoiseCov;//!< measurement noise covariance matrix (R)
2075 oclMat errorCovPre; //!< priori error estimate covariance matrix (P'(k)): P'(k)=A*P(k-1)*At + Q)*/
2076 oclMat gain; //!< Kalman gain matrix (K(k)): K(k)=P'(k)*Ht*inv(H*P'(k)*Ht+R)
2077 oclMat errorCovPost; //!< posteriori error estimate covariance matrix (P(k)): P(k)=(I-K(k)*H)*P'(k)
2086 /*!***************K Nearest Neighbour*************!*/
2087 class CV_EXPORTS KNearestNeighbour: public CvKNearest
2090 KNearestNeighbour();
2091 ~KNearestNeighbour();
2093 bool train(const Mat& trainData, Mat& labels, Mat& sampleIdx = Mat().setTo(Scalar::all(0)),
2094 bool isRegression = false, int max_k = 32, bool updateBase = false);
2098 void find_nearest(const oclMat& samples, int k, oclMat& lables);
2104 /*!*************** SVM *************!*/
2105 class CV_EXPORTS CvSVM_OCL : public CvSVM
2110 CvSVM_OCL(const cv::Mat& trainData, const cv::Mat& responses,
2111 const cv::Mat& varIdx=cv::Mat(), const cv::Mat& sampleIdx=cv::Mat(),
2112 CvSVMParams params=CvSVMParams());
2113 CV_WRAP float predict( const int row_index, Mat& src, bool returnDFVal=false ) const;
2114 CV_WRAP void predict( cv::InputArray samples, cv::OutputArray results ) const;
2115 CV_WRAP float predict( const cv::Mat& sample, bool returnDFVal=false ) const;
2116 float predict( const CvMat* samples, CV_OUT CvMat* results ) const;
2119 float predict( const int row_index, int row_len, Mat& src, bool returnDFVal=false ) const;
2120 void create_kernel();
2121 void create_solver();
2124 /*!*************** END *************!*/
2127 #if defined _MSC_VER && _MSC_VER >= 1200
2128 # pragma warning( push)
2129 # pragma warning( disable: 4267)
2131 #include "opencv2/ocl/matrix_operations.hpp"
2132 #if defined _MSC_VER && _MSC_VER >= 1200
2133 # pragma warning( pop)
2136 #endif /* __OPENCV_OCL_HPP__ */