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 oclMaterials 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"
60 CVCL_DEVICE_TYPE_DEFAULT = (1 << 0),
61 CVCL_DEVICE_TYPE_CPU = (1 << 1),
62 CVCL_DEVICE_TYPE_GPU = (1 << 2),
63 CVCL_DEVICE_TYPE_ACCELERATOR = (1 << 3),
64 //CVCL_DEVICE_TYPE_CUSTOM = (1 << 4)
65 CVCL_DEVICE_TYPE_ALL = 0xFFFFFFFF
77 DEVICE_MEM_DEFAULT = 0,
78 DEVICE_MEM_AHP, //alloc host pointer
79 DEVICE_MEM_UHP, //use host pointer
80 DEVICE_MEM_CHP, //copy host pointer
81 DEVICE_MEM_PM //persistent memory
84 //Get the global device memory and read/write type
85 //return 1 if unified memory system supported, otherwise return 0
86 CV_EXPORTS int getDevMemType(DevMemRW& rw_type, DevMemType& mem_type);
88 //Set the global device memory and read/write type,
89 //the newly generated oclMat will all use this type
90 //return -1 if the target type is unsupported, otherwise return 0
91 CV_EXPORTS int setDevMemType(DevMemRW rw_type = DEVICE_MEM_R_W, DevMemType mem_type = DEVICE_MEM_DEFAULT);
93 //this class contains ocl runtime information
104 Info &operator = (const Info &m);
105 std::vector<String> DeviceName;
108 //////////////////////////////// Initialization & Info ////////////////////////
109 //this function may be obsoleted
110 //CV_EXPORTS cl_device_id getDevice();
111 //the function must be called before any other cv::ocl::functions, it initialize ocl runtime
112 //each Info relates to an OpenCL platform
113 //there is one or more devices in each platform, each one has a separate name
114 CV_EXPORTS int getDevice(std::vector<Info> &oclinfo, int devicetype = CVCL_DEVICE_TYPE_GPU);
116 //set device you want to use, optional function after getDevice be called
117 //the devnum is the index of the selected device in DeviceName vector of INfo
118 CV_EXPORTS void setDevice(Info &oclinfo, int devnum = 0);
120 //The two functions below enable other opencl program to use ocl module's cl_context and cl_command_queue
121 //returns cl_context *
122 CV_EXPORTS void* getoclContext();
123 //returns cl_command_queue *
124 CV_EXPORTS void* getoclCommandQueue();
126 //explicit call clFinish. The global command queue will be used.
127 CV_EXPORTS void finish();
129 //this function enable ocl module to use customized cl_context and cl_command_queue
130 //getDevice also need to be called before this function
131 CV_EXPORTS void setDeviceEx(Info &oclinfo, void *ctx, void *qu, int devnum = 0);
133 //returns true when global OpenCL context is initialized
134 CV_EXPORTS bool initialized();
136 //////////////////////////////// OpenCL context ////////////////////////
137 //This is a global singleton class used to represent a OpenCL context.
138 class CV_EXPORTS Context
142 friend class std::auto_ptr<Context>;
143 friend bool initialized();
145 static std::auto_ptr<Context> clCxt;
152 static Context *getContext();
153 static void setContext(Info &oclinfo);
155 enum {CL_DOUBLE, CL_UNIFIED_MEM, CL_VER_1_2};
156 bool supportsFeature(int ftype);
157 size_t computeUnits();
158 size_t maxWorkGroupSize();
160 void* oclCommandQueue();
163 //! Calls a kernel, by string. Pass globalThreads = NULL, and cleanUp = true, to finally clean-up without executing.
164 CV_EXPORTS double openCLExecuteKernelInterop(Context *clCxt ,
165 const char **source, String kernelName,
166 size_t globalThreads[3], size_t localThreads[3],
167 std::vector< std::pair<size_t, const void *> > &args,
168 int channels, int depth, const char *build_options,
169 bool finish = true, bool measureKernelTime = false,
170 bool cleanUp = true);
172 //! Calls a kernel, by file. Pass globalThreads = NULL, and cleanUp = true, to finally clean-up without executing.
173 CV_EXPORTS double openCLExecuteKernelInterop(Context *clCxt ,
174 const char **fileName, const int numFiles, String kernelName,
175 size_t globalThreads[3], size_t localThreads[3],
176 std::vector< std::pair<size_t, const void *> > &args,
177 int channels, int depth, const char *build_options,
178 bool finish = true, bool measureKernelTime = false,
179 bool cleanUp = true);
181 //! Enable or disable OpenCL program binary caching onto local disk
182 // After a program (*.cl files in opencl/ folder) is built at runtime, we allow the
183 // compiled OpenCL program to be cached to the path automatically as "path/*.clb"
184 // binary file, which will be reused when the OpenCV executable is started again.
186 // Caching mode is controlled by the following enums
188 // 1. the feature is by default enabled when OpenCV is built in release mode.
189 // 2. the CACHE_DEBUG / CACHE_RELEASE flags only effectively work with MSVC compiler;
190 // for GNU compilers, the function always treats the build as release mode (enabled by default).
193 CACHE_NONE = 0, // do not cache OpenCL binary
194 CACHE_DEBUG = 0x1 << 0, // cache OpenCL binary when built in debug mode (only work with MSVC)
195 CACHE_RELEASE = 0x1 << 1, // default behavior, only cache when built in release mode (only work with MSVC)
196 CACHE_ALL = CACHE_DEBUG | CACHE_RELEASE, // always cache opencl binary
197 CACHE_UPDATE = 0x1 << 2 // if the binary cache file with the same name is already on the disk, it will be updated.
199 CV_EXPORTS void setBinaryDiskCache(int mode = CACHE_RELEASE, cv::String path = "./");
201 //! set where binary cache to be saved to
202 CV_EXPORTS void setBinpath(const char *path);
204 class CV_EXPORTS oclMatExpr;
205 //////////////////////////////// oclMat ////////////////////////////////
206 class CV_EXPORTS oclMat
209 //! default constructor
211 //! constructs oclMatrix of the specified size and type (_type is CV_8UC1, CV_64FC3, CV_32SC(12) etc.)
212 oclMat(int rows, int cols, int type);
213 oclMat(Size size, int type);
214 //! constucts oclMatrix and fills it with the specified value _s.
215 oclMat(int rows, int cols, int type, const Scalar &s);
216 oclMat(Size size, int type, const Scalar &s);
218 oclMat(const oclMat &m);
220 //! constructor for oclMatrix headers pointing to user-allocated data
221 oclMat(int rows, int cols, int type, void *data, size_t step = Mat::AUTO_STEP);
222 oclMat(Size size, int type, void *data, size_t step = Mat::AUTO_STEP);
224 //! creates a matrix header for a part of the bigger matrix
225 oclMat(const oclMat &m, const Range &rowRange, const Range &colRange);
226 oclMat(const oclMat &m, const Rect &roi);
228 //! builds oclMat from Mat. Perfom blocking upload to device.
229 explicit oclMat (const Mat &m);
231 //! destructor - calls release()
234 //! assignment operators
235 oclMat &operator = (const oclMat &m);
236 //! assignment operator. Perfom blocking upload to device.
237 oclMat &operator = (const Mat &m);
238 oclMat &operator = (const oclMatExpr& expr);
240 //! pefroms blocking upload data to oclMat.
241 void upload(const cv::Mat &m);
244 //! downloads data from device to host memory. Blocking calls.
245 operator Mat() const;
246 void download(cv::Mat &m) const;
248 //! convert to _InputArray
249 operator _InputArray();
251 //! convert to _OutputArray
252 operator _OutputArray();
254 //! returns a new oclMatrix header for the specified row
255 oclMat row(int y) const;
256 //! returns a new oclMatrix header for the specified column
257 oclMat col(int x) const;
258 //! ... for the specified row span
259 oclMat rowRange(int startrow, int endrow) const;
260 oclMat rowRange(const Range &r) const;
261 //! ... for the specified column span
262 oclMat colRange(int startcol, int endcol) const;
263 oclMat colRange(const Range &r) const;
265 //! returns deep copy of the oclMatrix, i.e. the data is copied
266 oclMat clone() const;
267 //! copies the oclMatrix content to "m".
268 // It calls m.create(this->size(), this->type()).
269 // It supports any data type
270 void copyTo( oclMat &m ) const;
271 //! copies those oclMatrix elements to "m" that are marked with non-zero mask elements.
272 //It supports 8UC1 8UC4 32SC1 32SC4 32FC1 32FC4
273 void copyTo( oclMat &m, const oclMat &mask ) const;
274 //! converts oclMatrix to another datatype with optional scalng. See cvConvertScale.
275 //It supports 8UC1 8UC4 32SC1 32SC4 32FC1 32FC4
276 void convertTo( oclMat &m, int rtype, double alpha = 1, double beta = 0 ) const;
278 void assignTo( oclMat &m, int type = -1 ) const;
280 //! sets every oclMatrix element to s
281 //It supports 8UC1 8UC4 32SC1 32SC4 32FC1 32FC4
282 oclMat& operator = (const Scalar &s);
283 //! sets some of the oclMatrix elements to s, according to the mask
284 //It supports 8UC1 8UC4 32SC1 32SC4 32FC1 32FC4
285 oclMat& setTo(const Scalar &s, const oclMat &mask = oclMat());
286 //! creates alternative oclMatrix header for the same data, with different
287 // number of channels and/or different number of rows. see cvReshape.
288 oclMat reshape(int cn, int rows = 0) const;
290 //! allocates new oclMatrix data unless the oclMatrix already has specified size and type.
291 // previous data is unreferenced if needed.
292 void create(int rows, int cols, int type);
293 void create(Size size, int type);
295 //! allocates new oclMatrix with specified device memory type.
296 void createEx(int rows, int cols, int type,
297 DevMemRW rw_type, DevMemType mem_type, void* hptr = 0);
298 void createEx(Size size, int type, DevMemRW rw_type,
299 DevMemType mem_type, void* hptr = 0);
301 //! decreases reference counter;
302 // deallocate the data when reference counter reaches 0.
305 //! swaps with other smart pointer
306 void swap(oclMat &mat);
308 //! locates oclMatrix header within a parent oclMatrix. See below
309 void locateROI( Size &wholeSize, Point &ofs ) const;
310 //! moves/resizes the current oclMatrix ROI inside the parent oclMatrix.
311 oclMat& adjustROI( int dtop, int dbottom, int dleft, int dright );
312 //! extracts a rectangular sub-oclMatrix
313 // (this is a generalized form of row, rowRange etc.)
314 oclMat operator()( Range rowRange, Range colRange ) const;
315 oclMat operator()( const Rect &roi ) const;
317 oclMat& operator+=( const oclMat& m );
318 oclMat& operator-=( const oclMat& m );
319 oclMat& operator*=( const oclMat& m );
320 oclMat& operator/=( const oclMat& m );
322 //! returns true if the oclMatrix data is continuous
323 // (i.e. when there are no gaps between successive rows).
324 // similar to CV_IS_oclMat_CONT(cvoclMat->type)
325 bool isContinuous() const;
326 //! returns element size in bytes,
327 // similar to CV_ELEM_SIZE(cvMat->type)
328 size_t elemSize() const;
329 //! returns the size of element channel in bytes.
330 size_t elemSize1() const;
331 //! returns element type, similar to CV_MAT_TYPE(cvMat->type)
333 //! returns element type, i.e. 8UC3 returns 8UC4 because in ocl
334 //! 3 channels element actually use 4 channel space
336 //! returns element type, similar to CV_MAT_DEPTH(cvMat->type)
338 //! returns element type, similar to CV_MAT_CN(cvMat->type)
339 int channels() const;
340 //! returns element type, return 4 for 3 channels element,
341 //!becuase 3 channels element actually use 4 channel space
342 int oclchannels() const;
343 //! returns step/elemSize1()
344 size_t step1() const;
345 //! returns oclMatrix size:
346 // width == number of columns, height == number of rows
348 //! returns true if oclMatrix data is NULL
351 //! returns pointer to y-th row
352 uchar* ptr(int y = 0);
353 const uchar *ptr(int y = 0) const;
355 //! template version of the above method
356 template<typename _Tp> _Tp *ptr(int y = 0);
357 template<typename _Tp> const _Tp *ptr(int y = 0) const;
359 //! matrix transposition
362 /*! includes several bit-fields:
363 - the magic signature
369 //! the number of rows and columns
371 //! a distance between successive rows in bytes; includes the gap if any
373 //! pointer to the data(OCL memory object)
376 //! pointer to the reference counter;
377 // when oclMatrix points to user-allocated data, the pointer is NULL
380 //! helper fields used in locateROI and adjustROI
381 //datastart and dataend are not used in current version
385 //! OpenCL context associated with the oclMat object.
387 //add offset for handle ROI, calculated in byte
389 //add wholerows and wholecols for the whole matrix, datastart and dataend are no longer used
394 // convert InputArray/OutputArray to oclMat references
395 CV_EXPORTS oclMat& getOclMatRef(InputArray src);
396 CV_EXPORTS oclMat& getOclMatRef(OutputArray src);
398 ///////////////////// mat split and merge /////////////////////////////////
399 //! Compose a multi-channel array from several single-channel arrays
401 CV_EXPORTS void merge(const oclMat *src, size_t n, oclMat &dst);
402 CV_EXPORTS void merge(const std::vector<oclMat> &src, oclMat &dst);
404 //! Divides multi-channel array into several single-channel arrays
406 CV_EXPORTS void split(const oclMat &src, oclMat *dst);
407 CV_EXPORTS void split(const oclMat &src, std::vector<oclMat> &dst);
409 ////////////////////////////// Arithmetics ///////////////////////////////////
410 //#if defined DOUBLE_SUPPORT
415 // CV_EXPORTS void addWeighted(const oclMat& a,F alpha, const oclMat& b,F beta,F gama, oclMat& c);
416 CV_EXPORTS void addWeighted(const oclMat &a, double alpha, const oclMat &b, double beta, double gama, oclMat &c);
417 //! adds one matrix to another (c = a + b)
418 // supports all types except CV_8SC1,CV_8SC2,CV8SC3 and CV_8SC4
419 CV_EXPORTS void add(const oclMat &a, const oclMat &b, oclMat &c);
420 //! adds one matrix to another (c = a + b)
421 // supports all types except CV_8SC1,CV_8SC2,CV8SC3 and CV_8SC4
422 CV_EXPORTS void add(const oclMat &a, const oclMat &b, oclMat &c, const oclMat &mask);
423 //! adds scalar to a matrix (c = a + s)
424 // supports all types except CV_8SC1,CV_8SC2,CV8SC3 and CV_8SC4
425 CV_EXPORTS void add(const oclMat &a, const Scalar &sc, oclMat &c, const oclMat &mask = oclMat());
426 //! subtracts one matrix from another (c = a - b)
427 // supports all types except CV_8SC1,CV_8SC2,CV8SC3 and CV_8SC4
428 CV_EXPORTS void subtract(const oclMat &a, const oclMat &b, oclMat &c);
429 //! subtracts one matrix from another (c = a - b)
430 // supports all types except CV_8SC1,CV_8SC2,CV8SC3 and CV_8SC4
431 CV_EXPORTS void subtract(const oclMat &a, const oclMat &b, oclMat &c, const oclMat &mask);
432 //! subtracts scalar from a matrix (c = a - s)
433 // supports all types except CV_8SC1,CV_8SC2,CV8SC3 and CV_8SC4
434 CV_EXPORTS void subtract(const oclMat &a, const Scalar &sc, oclMat &c, const oclMat &mask = oclMat());
435 //! subtracts scalar from a matrix (c = a - s)
436 // supports all types except CV_8SC1,CV_8SC2,CV8SC3 and CV_8SC4
437 CV_EXPORTS void subtract(const Scalar &sc, const oclMat &a, oclMat &c, const oclMat &mask = oclMat());
438 //! computes element-wise product of the two arrays (c = a * b)
439 // supports all types except CV_8SC1,CV_8SC2,CV8SC3 and CV_8SC4
440 CV_EXPORTS void multiply(const oclMat &a, const oclMat &b, oclMat &c, double scale = 1);
441 //! multiplies matrix to a number (dst = scalar * src)
442 // supports CV_32FC1 only
443 CV_EXPORTS void multiply(double scalar, const oclMat &src, oclMat &dst);
444 //! computes element-wise quotient of the two arrays (c = a / b)
445 // supports all types except CV_8SC1,CV_8SC2,CV8SC3 and CV_8SC4
446 CV_EXPORTS void divide(const oclMat &a, const oclMat &b, oclMat &c, double scale = 1);
447 //! computes element-wise quotient of the two arrays (c = a / b)
448 // supports all types except CV_8SC1,CV_8SC2,CV8SC3 and CV_8SC4
449 CV_EXPORTS void divide(double scale, const oclMat &b, oclMat &c);
451 //! compares elements of two arrays (c = a <cmpop> b)
452 // supports except CV_8SC1,CV_8SC2,CV8SC3,CV_8SC4 types
453 CV_EXPORTS void compare(const oclMat &a, const oclMat &b, oclMat &c, int cmpop);
455 //! transposes the matrix
456 // supports CV_8UC1, 8UC4, 8SC4, 16UC2, 16SC2, 32SC1 and 32FC1.(the same as cuda)
457 CV_EXPORTS void transpose(const oclMat &src, oclMat &dst);
459 //! computes element-wise absolute difference of two arrays (c = abs(a - b))
460 // supports all types except CV_8SC1,CV_8SC2,CV8SC3 and CV_8SC4
461 CV_EXPORTS void absdiff(const oclMat &a, const oclMat &b, oclMat &c);
462 //! computes element-wise absolute difference of array and scalar (c = abs(a - s))
463 // supports all types except CV_8SC1,CV_8SC2,CV8SC3 and CV_8SC4
464 CV_EXPORTS void absdiff(const oclMat &a, const Scalar &s, oclMat &c);
466 //! computes mean value and standard deviation of all or selected array elements
467 // supports except CV_32F,CV_64F
468 CV_EXPORTS void meanStdDev(const oclMat &mtx, Scalar &mean, Scalar &stddev);
470 //! computes norm of array
471 // supports NORM_INF, NORM_L1, NORM_L2
472 // supports only CV_8UC1 type
473 CV_EXPORTS double norm(const oclMat &src1, int normType = NORM_L2);
475 //! computes norm of the difference between two arrays
476 // supports NORM_INF, NORM_L1, NORM_L2
477 // supports only CV_8UC1 type
478 CV_EXPORTS double norm(const oclMat &src1, const oclMat &src2, int normType = NORM_L2);
480 //! reverses the order of the rows, columns or both in a matrix
481 // supports all types
482 CV_EXPORTS void flip(const oclMat &a, oclMat &b, int flipCode);
484 //! computes sum of array elements
485 // disabled until fix crash
487 CV_EXPORTS Scalar sum(const oclMat &m);
488 CV_EXPORTS Scalar absSum(const oclMat &m);
489 CV_EXPORTS Scalar sqrSum(const oclMat &m);
491 //! finds global minimum and maximum array elements and returns their values
492 // support all C1 types
494 CV_EXPORTS void minMax(const oclMat &src, double *minVal, double *maxVal = 0, const oclMat &mask = oclMat());
495 CV_EXPORTS void minMax_buf(const oclMat &src, double *minVal, double *maxVal, const oclMat &mask, oclMat& buf);
497 //! finds global minimum and maximum array elements and returns their values with locations
498 // support all C1 types
500 CV_EXPORTS void minMaxLoc(const oclMat &src, double *minVal, double *maxVal = 0, Point *minLoc = 0, Point *maxLoc = 0,
501 const oclMat &mask = oclMat());
503 //! counts non-zero array elements
505 CV_EXPORTS int countNonZero(const oclMat &src);
507 //! transforms 8-bit unsigned integers using lookup table: dst(i)=lut(src(i))
508 // destination array will have the depth type as lut and the same channels number as source
509 //It supports 8UC1 8UC4 only
510 CV_EXPORTS void LUT(const oclMat &src, const oclMat &lut, oclMat &dst);
512 //! only 8UC1 and 256 bins is supported now
513 CV_EXPORTS void calcHist(const oclMat &mat_src, oclMat &mat_hist);
514 //! only 8UC1 and 256 bins is supported now
515 CV_EXPORTS void equalizeHist(const oclMat &mat_src, oclMat &mat_dst);
517 //! only 8UC1 is supported now
518 CV_EXPORTS Ptr<cv::CLAHE> createCLAHE(double clipLimit = 40.0, Size tileGridSize = Size(8, 8));
521 // supports 8UC1 8UC4
522 CV_EXPORTS void bilateralFilter(const oclMat& src, oclMat& dst, int d, double sigmaColor, double sigmaSpace, int borderType=BORDER_DEFAULT);
524 //! Applies an adaptive bilateral filter to the input image
525 // This is not truly a bilateral filter. Instead of using user provided fixed parameters,
526 // the function calculates a constant at each window based on local standard deviation,
527 // and use this constant to do filtering.
528 // supports 8UC1 8UC3
529 CV_EXPORTS void adaptiveBilateralFilter(const oclMat& src, oclMat& dst, Size ksize, double sigmaSpace, Point anchor = Point(-1, -1), int borderType=BORDER_DEFAULT);
531 //! computes exponent of each matrix element (b = e**a)
532 // supports only CV_32FC1 type
533 CV_EXPORTS void exp(const oclMat &a, oclMat &b);
535 //! computes natural logarithm of absolute value of each matrix element: b = log(abs(a))
536 // supports only CV_32FC1 type
537 CV_EXPORTS void log(const oclMat &a, oclMat &b);
539 //! computes magnitude of each (x(i), y(i)) vector
540 // supports only CV_32F CV_64F type
541 CV_EXPORTS void magnitude(const oclMat &x, const oclMat &y, oclMat &magnitude);
542 CV_EXPORTS void magnitudeSqr(const oclMat &x, const oclMat &y, oclMat &magnitude);
544 CV_EXPORTS void magnitudeSqr(const oclMat &x, oclMat &magnitude);
546 //! computes angle (angle(i)) of each (x(i), y(i)) vector
547 // supports only CV_32F CV_64F type
548 CV_EXPORTS void phase(const oclMat &x, const oclMat &y, oclMat &angle, bool angleInDegrees = false);
550 //! the function raises every element of tne input array to p
551 //! support only CV_32F CV_64F type
552 CV_EXPORTS void pow(const oclMat &x, double p, oclMat &y);
554 //! converts Cartesian coordinates to polar
555 // supports only CV_32F CV_64F type
556 CV_EXPORTS void cartToPolar(const oclMat &x, const oclMat &y, oclMat &magnitude, oclMat &angle, bool angleInDegrees = false);
558 //! converts polar coordinates to Cartesian
559 // supports only CV_32F CV_64F type
560 CV_EXPORTS void polarToCart(const oclMat &magnitude, const oclMat &angle, oclMat &x, oclMat &y, bool angleInDegrees = false);
562 //! perfroms per-elements bit-wise inversion
563 // supports all types
564 CV_EXPORTS void bitwise_not(const oclMat &src, oclMat &dst);
565 //! calculates per-element bit-wise disjunction of two arrays
566 // supports all types
567 CV_EXPORTS void bitwise_or(const oclMat &src1, const oclMat &src2, oclMat &dst, const oclMat &mask = oclMat());
568 CV_EXPORTS void bitwise_or(const oclMat &src1, const Scalar &s, oclMat &dst, const oclMat &mask = oclMat());
569 //! calculates per-element bit-wise conjunction of two arrays
570 // supports all types
571 CV_EXPORTS void bitwise_and(const oclMat &src1, const oclMat &src2, oclMat &dst, const oclMat &mask = oclMat());
572 CV_EXPORTS void bitwise_and(const oclMat &src1, const Scalar &s, oclMat &dst, const oclMat &mask = oclMat());
573 //! calculates per-element bit-wise "exclusive or" operation
574 // supports all types
575 CV_EXPORTS void bitwise_xor(const oclMat &src1, const oclMat &src2, oclMat &dst, const oclMat &mask = oclMat());
576 CV_EXPORTS void bitwise_xor(const oclMat &src1, const Scalar &s, oclMat &dst, const oclMat &mask = oclMat());
578 //! Logical operators
579 CV_EXPORTS oclMat operator ~ (const oclMat &);
580 CV_EXPORTS oclMat operator | (const oclMat &, const oclMat &);
581 CV_EXPORTS oclMat operator & (const oclMat &, const oclMat &);
582 CV_EXPORTS oclMat operator ^ (const oclMat &, const oclMat &);
585 //! Mathematics operators
586 CV_EXPORTS oclMatExpr operator + (const oclMat &src1, const oclMat &src2);
587 CV_EXPORTS oclMatExpr operator - (const oclMat &src1, const oclMat &src2);
588 CV_EXPORTS oclMatExpr operator * (const oclMat &src1, const oclMat &src2);
589 CV_EXPORTS oclMatExpr operator / (const oclMat &src1, const oclMat &src2);
591 struct CV_EXPORTS ConvolveBuf
595 Size user_block_size;
598 oclMat image_spect, templ_spect, result_spect;
599 oclMat image_block, templ_block, result_data;
601 void create(Size image_size, Size templ_size);
602 static Size estimateBlockSize(Size result_size, Size templ_size);
605 //! computes convolution of two images, may use discrete Fourier transform
606 //! support only CV_32FC1 type
607 CV_EXPORTS void convolve(const oclMat &image, const oclMat &temp1, oclMat &result, bool ccorr = false);
608 CV_EXPORTS void convolve(const oclMat &image, const oclMat &temp1, oclMat &result, bool ccorr, ConvolveBuf& buf);
610 //! Performs a per-element multiplication of two Fourier spectrums.
611 //! Only full (not packed) CV_32FC2 complex spectrums in the interleaved format are supported for now.
612 //! support only CV_32FC2 type
613 CV_EXPORTS void mulSpectrums(const oclMat &a, const oclMat &b, oclMat &c, int flags, float scale, bool conjB = false);
615 CV_EXPORTS void cvtColor(const oclMat &src, oclMat &dst, int code , int dcn = 0);
617 //////////////////////////////// Filter Engine ////////////////////////////////
620 The Base Class for 1D or Row-wise Filters
622 This is the base class for linear or non-linear filters that process 1D data.
623 In particular, such filters are used for the "horizontal" filtering parts in separable filters.
625 class CV_EXPORTS BaseRowFilter_GPU
628 BaseRowFilter_GPU(int ksize_, int anchor_, int bordertype_) : ksize(ksize_), anchor(anchor_), bordertype(bordertype_) {}
629 virtual ~BaseRowFilter_GPU() {}
630 virtual void operator()(const oclMat &src, oclMat &dst) = 0;
631 int ksize, anchor, bordertype;
635 The Base Class for Column-wise Filters
637 This is the base class for linear or non-linear filters that process columns of 2D arrays.
638 Such filters are used for the "vertical" filtering parts in separable filters.
640 class CV_EXPORTS BaseColumnFilter_GPU
643 BaseColumnFilter_GPU(int ksize_, int anchor_, int bordertype_) : ksize(ksize_), anchor(anchor_), bordertype(bordertype_) {}
644 virtual ~BaseColumnFilter_GPU() {}
645 virtual void operator()(const oclMat &src, oclMat &dst) = 0;
646 int ksize, anchor, bordertype;
650 The Base Class for Non-Separable 2D Filters.
652 This is the base class for linear or non-linear 2D filters.
654 class CV_EXPORTS BaseFilter_GPU
657 BaseFilter_GPU(const Size &ksize_, const Point &anchor_, const int &borderType_)
658 : ksize(ksize_), anchor(anchor_), borderType(borderType_) {}
659 virtual ~BaseFilter_GPU() {}
660 virtual void operator()(const oclMat &src, oclMat &dst) = 0;
667 The Base Class for Filter Engine.
669 The class can be used to apply an arbitrary filtering operation to an image.
670 It contains all the necessary intermediate buffers.
672 class CV_EXPORTS FilterEngine_GPU
675 virtual ~FilterEngine_GPU() {}
677 virtual void apply(const oclMat &src, oclMat &dst, Rect roi = Rect(0, 0, -1, -1)) = 0;
680 //! returns the non-separable filter engine with the specified filter
681 CV_EXPORTS Ptr<FilterEngine_GPU> createFilter2D_GPU(const Ptr<BaseFilter_GPU> filter2D);
683 //! returns the primitive row filter with the specified kernel
684 CV_EXPORTS Ptr<BaseRowFilter_GPU> getLinearRowFilter_GPU(int srcType, int bufType, const Mat &rowKernel,
685 int anchor = -1, int bordertype = BORDER_DEFAULT);
687 //! returns the primitive column filter with the specified kernel
688 CV_EXPORTS Ptr<BaseColumnFilter_GPU> getLinearColumnFilter_GPU(int bufType, int dstType, const Mat &columnKernel,
689 int anchor = -1, int bordertype = BORDER_DEFAULT, double delta = 0.0);
691 //! returns the separable linear filter engine
692 CV_EXPORTS Ptr<FilterEngine_GPU> createSeparableLinearFilter_GPU(int srcType, int dstType, const Mat &rowKernel,
693 const Mat &columnKernel, const Point &anchor = Point(-1, -1), double delta = 0.0, int bordertype = BORDER_DEFAULT);
695 //! returns the separable filter engine with the specified filters
696 CV_EXPORTS Ptr<FilterEngine_GPU> createSeparableFilter_GPU(const Ptr<BaseRowFilter_GPU> &rowFilter,
697 const Ptr<BaseColumnFilter_GPU> &columnFilter);
699 //! returns the Gaussian filter engine
700 CV_EXPORTS Ptr<FilterEngine_GPU> createGaussianFilter_GPU(int type, Size ksize, double sigma1, double sigma2 = 0, int bordertype = BORDER_DEFAULT);
702 //! returns filter engine for the generalized Sobel operator
703 CV_EXPORTS Ptr<FilterEngine_GPU> createDerivFilter_GPU( int srcType, int dstType, int dx, int dy, int ksize, int borderType = BORDER_DEFAULT );
705 //! applies Laplacian operator to the image
706 // supports only ksize = 1 and ksize = 3 8UC1 8UC4 32FC1 32FC4 data type
707 CV_EXPORTS void Laplacian(const oclMat &src, oclMat &dst, int ddepth, int ksize = 1, double scale = 1);
709 //! returns 2D box filter
710 // supports CV_8UC1 and CV_8UC4 source type, dst type must be the same as source type
711 CV_EXPORTS Ptr<BaseFilter_GPU> getBoxFilter_GPU(int srcType, int dstType,
712 const Size &ksize, Point anchor = Point(-1, -1), int borderType = BORDER_DEFAULT);
714 //! returns box filter engine
715 CV_EXPORTS Ptr<FilterEngine_GPU> createBoxFilter_GPU(int srcType, int dstType, const Size &ksize,
716 const Point &anchor = Point(-1, -1), int borderType = BORDER_DEFAULT);
718 //! returns 2D filter with the specified kernel
719 // supports CV_8UC1 and CV_8UC4 types
720 CV_EXPORTS Ptr<BaseFilter_GPU> getLinearFilter_GPU(int srcType, int dstType, const Mat &kernel, const Size &ksize,
721 const Point &anchor = Point(-1, -1), int borderType = BORDER_DEFAULT);
723 //! returns the non-separable linear filter engine
724 CV_EXPORTS Ptr<FilterEngine_GPU> createLinearFilter_GPU(int srcType, int dstType, const Mat &kernel,
725 const Point &anchor = Point(-1, -1), int borderType = BORDER_DEFAULT);
727 //! smooths the image using the normalized box filter
728 // supports data type: CV_8UC1, CV_8UC4, CV_32FC1 and CV_32FC4
729 // supports border type: BORDER_CONSTANT, BORDER_REPLICATE, BORDER_REFLECT,BORDER_REFLECT_101,BORDER_WRAP
730 CV_EXPORTS void boxFilter(const oclMat &src, oclMat &dst, int ddepth, Size ksize,
731 Point anchor = Point(-1, -1), int borderType = BORDER_DEFAULT);
733 //! returns 2D morphological filter
734 //! only MORPH_ERODE and MORPH_DILATE are supported
735 // supports CV_8UC1, CV_8UC4, CV_32FC1 and CV_32FC4 types
736 // kernel must have CV_8UC1 type, one rows and cols == ksize.width * ksize.height
737 CV_EXPORTS Ptr<BaseFilter_GPU> getMorphologyFilter_GPU(int op, int type, const Mat &kernel, const Size &ksize,
738 Point anchor = Point(-1, -1));
740 //! returns morphological filter engine. Only MORPH_ERODE and MORPH_DILATE are supported.
741 CV_EXPORTS Ptr<FilterEngine_GPU> createMorphologyFilter_GPU(int op, int type, const Mat &kernel,
742 const Point &anchor = Point(-1, -1), int iterations = 1);
744 //! a synonym for normalized box filter
745 // supports data type: CV_8UC1, CV_8UC4, CV_32FC1 and CV_32FC4
746 // supports border type: BORDER_CONSTANT, BORDER_REPLICATE, BORDER_REFLECT,BORDER_REFLECT_101
747 static inline void blur(const oclMat &src, oclMat &dst, Size ksize, Point anchor = Point(-1, -1),
748 int borderType = BORDER_CONSTANT)
750 boxFilter(src, dst, -1, ksize, anchor, borderType);
753 //! applies non-separable 2D linear filter to the image
754 // Note, at the moment this function only works when anchor point is in the kernel center
755 // and kernel size supported is either 3x3 or 5x5; otherwise the function will fail to output valid result
756 CV_EXPORTS void filter2D(const oclMat &src, oclMat &dst, int ddepth, const Mat &kernel,
757 Point anchor = Point(-1, -1), int borderType = BORDER_DEFAULT);
759 //! applies separable 2D linear filter to the image
760 CV_EXPORTS void sepFilter2D(const oclMat &src, oclMat &dst, int ddepth, const Mat &kernelX, const Mat &kernelY,
761 Point anchor = Point(-1, -1), double delta = 0.0, int bordertype = BORDER_DEFAULT);
763 //! applies generalized Sobel operator to the image
764 // dst.type must equalize src.type
765 // supports data type: CV_8UC1, CV_8UC4, CV_32FC1 and CV_32FC4
766 // supports border type: BORDER_CONSTANT, BORDER_REPLICATE, BORDER_REFLECT,BORDER_REFLECT_101
767 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);
769 //! applies the vertical or horizontal Scharr operator to the image
770 // dst.type must equalize src.type
771 // supports data type: CV_8UC1, CV_8UC4, CV_32FC1 and CV_32FC4
772 // supports border type: BORDER_CONSTANT, BORDER_REPLICATE, BORDER_REFLECT,BORDER_REFLECT_101
773 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);
775 //! smooths the image using Gaussian filter.
776 // dst.type must equalize src.type
777 // supports data type: CV_8UC1, CV_8UC4, CV_32FC1 and CV_32FC4
778 // supports border type: BORDER_CONSTANT, BORDER_REPLICATE, BORDER_REFLECT,BORDER_REFLECT_101
779 CV_EXPORTS void GaussianBlur(const oclMat &src, oclMat &dst, Size ksize, double sigma1, double sigma2 = 0, int bordertype = BORDER_DEFAULT);
781 //! erodes the image (applies the local minimum operator)
782 // supports data type: CV_8UC1, CV_8UC4, CV_32FC1 and CV_32FC4
783 CV_EXPORTS void erode( const oclMat &src, oclMat &dst, const Mat &kernel, Point anchor = Point(-1, -1), int iterations = 1,
785 int borderType = BORDER_CONSTANT, const Scalar &borderValue = morphologyDefaultBorderValue());
788 //! dilates the image (applies the local maximum operator)
789 // supports data type: CV_8UC1, CV_8UC4, CV_32FC1 and CV_32FC4
790 CV_EXPORTS void dilate( const oclMat &src, oclMat &dst, const Mat &kernel, Point anchor = Point(-1, -1), int iterations = 1,
792 int borderType = BORDER_CONSTANT, const Scalar &borderValue = morphologyDefaultBorderValue());
795 //! applies an advanced morphological operation to the image
796 CV_EXPORTS void morphologyEx( const oclMat &src, oclMat &dst, int op, const Mat &kernel, Point anchor = Point(-1, -1), int iterations = 1,
798 int borderType = BORDER_CONSTANT, const Scalar &borderValue = morphologyDefaultBorderValue());
801 ////////////////////////////// Image processing //////////////////////////////
802 //! Does mean shift filtering on GPU.
803 CV_EXPORTS void meanShiftFiltering(const oclMat &src, oclMat &dst, int sp, int sr,
804 TermCriteria criteria = TermCriteria(TermCriteria::MAX_ITER + TermCriteria::EPS, 5, 1));
806 //! Does mean shift procedure on GPU.
807 CV_EXPORTS void meanShiftProc(const oclMat &src, oclMat &dstr, oclMat &dstsp, int sp, int sr,
808 TermCriteria criteria = TermCriteria(TermCriteria::MAX_ITER + TermCriteria::EPS, 5, 1));
810 //! Does mean shift segmentation with elimiation of small regions.
811 CV_EXPORTS void meanShiftSegmentation(const oclMat &src, Mat &dst, int sp, int sr, int minsize,
812 TermCriteria criteria = TermCriteria(TermCriteria::MAX_ITER + TermCriteria::EPS, 5, 1));
814 //! applies fixed threshold to the image.
815 // supports CV_8UC1 and CV_32FC1 data type
816 // supports threshold type: THRESH_BINARY, THRESH_BINARY_INV, THRESH_TRUNC, THRESH_TOZERO, THRESH_TOZERO_INV
817 CV_EXPORTS double threshold(const oclMat &src, oclMat &dst, double thresh, double maxVal, int type = THRESH_TRUNC);
819 //! resizes the image
820 // Supports INTER_NEAREST, INTER_LINEAR
821 // supports CV_8UC1, CV_8UC4, CV_32FC1 and CV_32FC4 types
822 CV_EXPORTS void resize(const oclMat &src, oclMat &dst, Size dsize, double fx = 0, double fy = 0, int interpolation = INTER_LINEAR);
824 //! Applies a generic geometrical transformation to an image.
826 // Supports INTER_NEAREST, INTER_LINEAR.
828 // Map1 supports CV_16SC2, CV_32FC2 types.
830 // Src supports CV_8UC1, CV_8UC2, CV_8UC4.
832 CV_EXPORTS void remap(const oclMat &src, oclMat &dst, oclMat &map1, oclMat &map2, int interpolation, int bordertype, const Scalar &value = Scalar());
834 //! copies 2D array to a larger destination array and pads borders with user-specifiable constant
835 // supports CV_8UC1, CV_8UC4, CV_32SC1 types
836 CV_EXPORTS void copyMakeBorder(const oclMat &src, oclMat &dst, int top, int bottom, int left, int right, int boardtype, const Scalar &value = Scalar());
838 //! Smoothes image using median filter
839 // The source 1- or 4-channel image. When m is 3 or 5, the image depth should be CV 8U or CV 32F.
840 CV_EXPORTS void medianFilter(const oclMat &src, oclMat &dst, int m);
842 //! warps the image using affine transformation
843 // Supports INTER_NEAREST, INTER_LINEAR, INTER_CUBIC
844 // supports CV_8UC1, CV_8UC4, CV_32FC1 and CV_32FC4 types
845 CV_EXPORTS void warpAffine(const oclMat &src, oclMat &dst, const Mat &M, Size dsize, int flags = INTER_LINEAR);
847 //! warps the image using perspective transformation
848 // Supports INTER_NEAREST, INTER_LINEAR, INTER_CUBIC
849 // supports CV_8UC1, CV_8UC4, CV_32FC1 and CV_32FC4 types
850 CV_EXPORTS void warpPerspective(const oclMat &src, oclMat &dst, const Mat &M, Size dsize, int flags = INTER_LINEAR);
852 //! computes the integral image and integral for the squared image
853 // sum will have CV_32S type, sqsum - CV32F type
854 // supports only CV_8UC1 source type
855 CV_EXPORTS void integral(const oclMat &src, oclMat &sum, oclMat &sqsum);
856 CV_EXPORTS void integral(const oclMat &src, oclMat &sum);
857 CV_EXPORTS void cornerHarris(const oclMat &src, oclMat &dst, int blockSize, int ksize, double k, int bordertype = cv::BORDER_DEFAULT);
858 CV_EXPORTS void cornerHarris_dxdy(const oclMat &src, oclMat &dst, oclMat &Dx, oclMat &Dy,
859 int blockSize, int ksize, double k, int bordertype = cv::BORDER_DEFAULT);
860 CV_EXPORTS void cornerMinEigenVal(const oclMat &src, oclMat &dst, int blockSize, int ksize, int bordertype = cv::BORDER_DEFAULT);
861 CV_EXPORTS void cornerMinEigenVal_dxdy(const oclMat &src, oclMat &dst, oclMat &Dx, oclMat &Dy,
862 int blockSize, int ksize, int bordertype = cv::BORDER_DEFAULT);
865 /////////////////////////////////// ML ///////////////////////////////////////////
867 //! Compute closest centers for each lines in source and lable it after center's index
868 // supports CV_32FC1/CV_32FC2/CV_32FC4 data type
869 CV_EXPORTS void distanceToCenters(oclMat &dists, oclMat &labels, const oclMat &src, const oclMat ¢ers);
871 //!Does k-means procedure on GPU
872 // supports CV_32FC1/CV_32FC2/CV_32FC4 data type
873 CV_EXPORTS double kmeans(const oclMat &src, int K, oclMat &bestLabels,
874 TermCriteria criteria, int attemps, int flags, oclMat ¢ers);
877 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
878 ///////////////////////////////////////////CascadeClassifier//////////////////////////////////////////////////////////////////
879 ///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
880 class CV_EXPORTS OclCascadeClassifier : public cv::CascadeClassifier
883 void detectMultiScale(oclMat &image, CV_OUT std::vector<cv::Rect>& faces,
884 double scaleFactor = 1.1, int minNeighbors = 3, int flags = 0,
885 Size minSize = Size(), Size maxSize = Size());
888 /////////////////////////////// Pyramid /////////////////////////////////////
889 CV_EXPORTS void pyrDown(const oclMat &src, oclMat &dst);
891 //! upsamples the source image and then smoothes it
892 CV_EXPORTS void pyrUp(const oclMat &src, oclMat &dst);
894 //! performs linear blending of two images
895 //! to avoid accuracy errors sum of weigths shouldn't be very close to zero
896 // supports only CV_8UC1 source type
897 CV_EXPORTS void blendLinear(const oclMat &img1, const oclMat &img2, const oclMat &weights1, const oclMat &weights2, oclMat &result);
899 //! computes vertical sum, supports only CV_32FC1 images
900 CV_EXPORTS void columnSum(const oclMat &src, oclMat &sum);
902 ///////////////////////////////////////// match_template /////////////////////////////////////////////////////////////
903 struct CV_EXPORTS MatchTemplateBuf
905 Size user_block_size;
906 oclMat imagef, templf;
907 std::vector<oclMat> images;
908 std::vector<oclMat> image_sums;
909 std::vector<oclMat> image_sqsums;
912 //! computes the proximity map for the raster template and the image where the template is searched for
913 // Supports TM_SQDIFF, TM_SQDIFF_NORMED, TM_CCORR, TM_CCORR_NORMED, TM_CCOEFF, TM_CCOEFF_NORMED for type 8UC1 and 8UC4
914 // Supports TM_SQDIFF, TM_CCORR for type 32FC1 and 32FC4
915 CV_EXPORTS void matchTemplate(const oclMat &image, const oclMat &templ, oclMat &result, int method);
917 //! computes the proximity map for the raster template and the image where the template is searched for
918 // Supports TM_SQDIFF, TM_SQDIFF_NORMED, TM_CCORR, TM_CCORR_NORMED, TM_CCOEFF, TM_CCOEFF_NORMED for type 8UC1 and 8UC4
919 // Supports TM_SQDIFF, TM_CCORR for type 32FC1 and 32FC4
920 CV_EXPORTS void matchTemplate(const oclMat &image, const oclMat &templ, oclMat &result, int method, MatchTemplateBuf &buf);
924 ///////////////////////////////////////////// Canny /////////////////////////////////////////////
925 struct CV_EXPORTS CannyBuf;
927 //! compute edges of the input image using Canny operator
928 // Support CV_8UC1 only
929 CV_EXPORTS void Canny(const oclMat &image, oclMat &edges, double low_thresh, double high_thresh, int apperture_size = 3, bool L2gradient = false);
930 CV_EXPORTS void Canny(const oclMat &image, CannyBuf &buf, oclMat &edges, double low_thresh, double high_thresh, int apperture_size = 3, bool L2gradient = false);
931 CV_EXPORTS void Canny(const oclMat &dx, const oclMat &dy, oclMat &edges, double low_thresh, double high_thresh, bool L2gradient = false);
932 CV_EXPORTS void Canny(const oclMat &dx, const oclMat &dy, CannyBuf &buf, oclMat &edges, double low_thresh, double high_thresh, bool L2gradient = false);
934 struct CV_EXPORTS CannyBuf
936 CannyBuf() : counter(NULL) {}
941 explicit CannyBuf(const Size &image_size, int apperture_size = 3) : counter(NULL)
943 create(image_size, apperture_size);
945 CannyBuf(const oclMat &dx_, const oclMat &dy_);
946 void create(const Size &image_size, int apperture_size = 3);
950 oclMat dx_buf, dy_buf;
951 oclMat magBuf, mapBuf;
952 oclMat trackBuf1, trackBuf2;
954 Ptr<FilterEngine_GPU> filterDX, filterDY;
957 ///////////////////////////////////////// Hough Transform /////////////////////////////////////////
959 struct HoughCirclesBuf
968 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);
969 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);
970 CV_EXPORTS void HoughCirclesDownload(const oclMat& d_circles, OutputArray h_circles);
973 ///////////////////////////////////////// clAmdFft related /////////////////////////////////////////
974 //! Performs a forward or inverse discrete Fourier transform (1D or 2D) of floating point matrix.
975 //! Param dft_size is the size of DFT transform.
977 //! For complex-to-real transform it is assumed that the source matrix is packed in CLFFT's format.
978 // support src type of CV32FC1, CV32FC2
979 // support flags: DFT_INVERSE, DFT_REAL_OUTPUT, DFT_COMPLEX_OUTPUT, DFT_ROWS
980 // dft_size is the size of original input, which is used for transformation from complex to real.
981 // dft_size must be powers of 2, 3 and 5
982 // real to complex dft requires at least v1.8 clAmdFft
983 // real to complex dft output is not the same with cpu version
984 // real to complex and complex to real does not support DFT_ROWS
985 CV_EXPORTS void dft(const oclMat &src, oclMat &dst, Size dft_size = Size(0, 0), int flags = 0);
987 //! implements generalized matrix product algorithm GEMM from BLAS
988 // The functionality requires clAmdBlas library
989 // only support type CV_32FC1
990 // flag GEMM_3_T is not supported
991 CV_EXPORTS void gemm(const oclMat &src1, const oclMat &src2, double alpha,
992 const oclMat &src3, double beta, oclMat &dst, int flags = 0);
994 //////////////// HOG (Histogram-of-Oriented-Gradients) Descriptor and Object Detector //////////////
996 struct CV_EXPORTS HOGDescriptor
1000 enum { DEFAULT_WIN_SIGMA = -1 };
1002 enum { DEFAULT_NLEVELS = 64 };
1004 enum { DESCR_FORMAT_ROW_BY_ROW, DESCR_FORMAT_COL_BY_COL };
1008 HOGDescriptor(Size win_size = Size(64, 128), Size block_size = Size(16, 16),
1010 Size block_stride = Size(8, 8), Size cell_size = Size(8, 8),
1012 int nbins = 9, double win_sigma = DEFAULT_WIN_SIGMA,
1014 double threshold_L2hys = 0.2, bool gamma_correction = true,
1016 int nlevels = DEFAULT_NLEVELS);
1020 size_t getDescriptorSize() const;
1022 size_t getBlockHistogramSize() const;
1026 void setSVMDetector(const std::vector<float> &detector);
1030 static std::vector<float> getDefaultPeopleDetector();
1032 static std::vector<float> getPeopleDetector48x96();
1034 static std::vector<float> getPeopleDetector64x128();
1038 void detect(const oclMat &img, std::vector<Point> &found_locations,
1040 double hit_threshold = 0, Size win_stride = Size(),
1042 Size padding = Size());
1046 void detectMultiScale(const oclMat &img, std::vector<Rect> &found_locations,
1048 double hit_threshold = 0, Size win_stride = Size(),
1050 Size padding = Size(), double scale0 = 1.05,
1052 int group_threshold = 2);
1056 void getDescriptors(const oclMat &img, Size win_stride,
1058 oclMat &descriptors,
1060 int descr_format = DESCR_FORMAT_COL_BY_COL);
1076 double threshold_L2hys;
1078 bool gamma_correction;
1086 // initialize buffers; only need to do once in case of multiscale detection
1088 void init_buffer(const oclMat &img, Size win_stride);
1092 void computeBlockHistograms(const oclMat &img);
1094 void computeGradient(const oclMat &img, oclMat &grad, oclMat &qangle);
1098 double getWinSigma() const;
1100 bool checkDetectorSize() const;
1104 static int numPartsWithin(int size, int part_size, int stride);
1106 static Size numPartsWithin(Size size, Size part_size, Size stride);
1110 // Coefficients of the separating plane
1118 // Results of the last classification step
1126 // Results of the last histogram evaluation step
1132 // Gradients conputation results
1134 oclMat grad, qangle;
1144 // effect size of input image (might be different from original size after scaling)
1151 ////////////////////////feature2d_ocl/////////////////
1152 /****************************************************************************************\
1154 \****************************************************************************************/
1155 template<typename T>
1156 struct CV_EXPORTS Accumulator
1160 template<> struct Accumulator<unsigned char>
1164 template<> struct Accumulator<unsigned short>
1168 template<> struct Accumulator<char>
1172 template<> struct Accumulator<short>
1178 * Manhattan distance (city block distance) functor
1181 struct CV_EXPORTS L1
1183 enum { normType = NORM_L1 };
1184 typedef T ValueType;
1185 typedef typename Accumulator<T>::Type ResultType;
1187 ResultType operator()( const T *a, const T *b, int size ) const
1189 return normL1<ValueType, ResultType>(a, b, size);
1194 * Euclidean distance functor
1197 struct CV_EXPORTS L2
1199 enum { normType = NORM_L2 };
1200 typedef T ValueType;
1201 typedef typename Accumulator<T>::Type ResultType;
1203 ResultType operator()( const T *a, const T *b, int size ) const
1205 return (ResultType)std::sqrt((double)normL2Sqr<ValueType, ResultType>(a, b, size));
1210 * Hamming distance functor - counts the bit differences between two strings - useful for the Brief descriptor
1211 * bit count of A exclusive XOR'ed with B
1213 struct CV_EXPORTS Hamming
1215 enum { normType = NORM_HAMMING };
1216 typedef unsigned char ValueType;
1217 typedef int ResultType;
1219 /** this will count the bits in a ^ b
1221 ResultType operator()( const unsigned char *a, const unsigned char *b, int size ) const
1223 return normHamming(a, b, size);
1227 ////////////////////////////////// BruteForceMatcher //////////////////////////////////
1229 class CV_EXPORTS BruteForceMatcher_OCL_base
1232 enum DistType {L1Dist = 0, L2Dist, HammingDist};
1233 explicit BruteForceMatcher_OCL_base(DistType distType = L2Dist);
1235 // Add descriptors to train descriptor collection
1236 void add(const std::vector<oclMat> &descCollection);
1238 // Get train descriptors collection
1239 const std::vector<oclMat> &getTrainDescriptors() const;
1241 // Clear train descriptors collection
1244 // Return true if there are not train descriptors in collection
1247 // Return true if the matcher supports mask in match methods
1248 bool isMaskSupported() const;
1250 // Find one best match for each query descriptor
1251 void matchSingle(const oclMat &query, const oclMat &train,
1252 oclMat &trainIdx, oclMat &distance,
1253 const oclMat &mask = oclMat());
1255 // Download trainIdx and distance and convert it to CPU vector with DMatch
1256 static void matchDownload(const oclMat &trainIdx, const oclMat &distance, std::vector<DMatch> &matches);
1257 // Convert trainIdx and distance to vector with DMatch
1258 static void matchConvert(const Mat &trainIdx, const Mat &distance, std::vector<DMatch> &matches);
1260 // Find one best match for each query descriptor
1261 void match(const oclMat &query, const oclMat &train, std::vector<DMatch> &matches, const oclMat &mask = oclMat());
1263 // Make gpu collection of trains and masks in suitable format for matchCollection function
1264 void makeGpuCollection(oclMat &trainCollection, oclMat &maskCollection, const std::vector<oclMat> &masks = std::vector<oclMat>());
1266 // Find one best match from train collection for each query descriptor
1267 void matchCollection(const oclMat &query, const oclMat &trainCollection,
1268 oclMat &trainIdx, oclMat &imgIdx, oclMat &distance,
1269 const oclMat &masks = oclMat());
1271 // Download trainIdx, imgIdx and distance and convert it to vector with DMatch
1272 static void matchDownload(const oclMat &trainIdx, const oclMat &imgIdx, const oclMat &distance, std::vector<DMatch> &matches);
1273 // Convert trainIdx, imgIdx and distance to vector with DMatch
1274 static void matchConvert(const Mat &trainIdx, const Mat &imgIdx, const Mat &distance, std::vector<DMatch> &matches);
1276 // Find one best match from train collection for each query descriptor.
1277 void match(const oclMat &query, std::vector<DMatch> &matches, const std::vector<oclMat> &masks = std::vector<oclMat>());
1279 // Find k best matches for each query descriptor (in increasing order of distances)
1280 void knnMatchSingle(const oclMat &query, const oclMat &train,
1281 oclMat &trainIdx, oclMat &distance, oclMat &allDist, int k,
1282 const oclMat &mask = oclMat());
1284 // Download trainIdx and distance and convert it to vector with DMatch
1285 // compactResult is used when mask is not empty. If compactResult is false matches
1286 // vector will have the same size as queryDescriptors rows. If compactResult is true
1287 // matches vector will not contain matches for fully masked out query descriptors.
1288 static void knnMatchDownload(const oclMat &trainIdx, const oclMat &distance,
1289 std::vector< std::vector<DMatch> > &matches, bool compactResult = false);
1290 // Convert trainIdx and distance to vector with DMatch
1291 static void knnMatchConvert(const Mat &trainIdx, const Mat &distance,
1292 std::vector< std::vector<DMatch> > &matches, bool compactResult = false);
1294 // Find k best matches for each query descriptor (in increasing order of distances).
1295 // compactResult is used when mask is not empty. If compactResult is false matches
1296 // vector will have the same size as queryDescriptors rows. If compactResult is true
1297 // matches vector will not contain matches for fully masked out query descriptors.
1298 void knnMatch(const oclMat &query, const oclMat &train,
1299 std::vector< std::vector<DMatch> > &matches, int k, const oclMat &mask = oclMat(),
1300 bool compactResult = false);
1302 // Find k best matches from train collection for each query descriptor (in increasing order of distances)
1303 void knnMatch2Collection(const oclMat &query, const oclMat &trainCollection,
1304 oclMat &trainIdx, oclMat &imgIdx, oclMat &distance,
1305 const oclMat &maskCollection = oclMat());
1307 // Download trainIdx and distance and convert it to vector with DMatch
1308 // compactResult is used when mask is not empty. If compactResult is false matches
1309 // vector will have the same size as queryDescriptors rows. If compactResult is true
1310 // matches vector will not contain matches for fully masked out query descriptors.
1311 static void knnMatch2Download(const oclMat &trainIdx, const oclMat &imgIdx, const oclMat &distance,
1312 std::vector< std::vector<DMatch> > &matches, bool compactResult = false);
1313 // Convert trainIdx and distance to vector with DMatch
1314 static void knnMatch2Convert(const Mat &trainIdx, const Mat &imgIdx, const Mat &distance,
1315 std::vector< std::vector<DMatch> > &matches, bool compactResult = false);
1317 // Find k best matches for each query descriptor (in increasing order of distances).
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 void knnMatch(const oclMat &query, std::vector< std::vector<DMatch> > &matches, int k,
1322 const std::vector<oclMat> &masks = std::vector<oclMat>(), bool compactResult = false);
1324 // Find best matches for each query descriptor which have distance less than maxDistance.
1325 // nMatches.at<int>(0, queryIdx) will contain matches count for queryIdx.
1326 // carefully nMatches can be greater than trainIdx.cols - it means that matcher didn't find all matches,
1327 // because it didn't have enough memory.
1328 // If trainIdx is empty, then trainIdx and distance will be created with size nQuery x max((nTrain / 100), 10),
1329 // otherwize user can pass own allocated trainIdx and distance with size nQuery x nMaxMatches
1330 // Matches doesn't sorted.
1331 void radiusMatchSingle(const oclMat &query, const oclMat &train,
1332 oclMat &trainIdx, oclMat &distance, oclMat &nMatches, float maxDistance,
1333 const oclMat &mask = oclMat());
1335 // Download trainIdx, nMatches and distance and convert it to vector with DMatch.
1336 // matches will be sorted in increasing order of distances.
1337 // compactResult is used when mask is not empty. If compactResult is false matches
1338 // vector will have the same size as queryDescriptors rows. If compactResult is true
1339 // matches vector will not contain matches for fully masked out query descriptors.
1340 static void radiusMatchDownload(const oclMat &trainIdx, const oclMat &distance, const oclMat &nMatches,
1341 std::vector< std::vector<DMatch> > &matches, bool compactResult = false);
1342 // Convert trainIdx, nMatches and distance to vector with DMatch.
1343 static void radiusMatchConvert(const Mat &trainIdx, const Mat &distance, const Mat &nMatches,
1344 std::vector< std::vector<DMatch> > &matches, bool compactResult = false);
1346 // Find best matches for each query descriptor which have distance less than maxDistance
1347 // in increasing order of distances).
1348 void radiusMatch(const oclMat &query, const oclMat &train,
1349 std::vector< std::vector<DMatch> > &matches, float maxDistance,
1350 const oclMat &mask = oclMat(), bool compactResult = false);
1352 // Find best matches for each query descriptor which have distance less than maxDistance.
1353 // If trainIdx is empty, then trainIdx and distance will be created with size nQuery x max((nQuery / 100), 10),
1354 // otherwize user can pass own allocated trainIdx and distance with size nQuery x nMaxMatches
1355 // Matches doesn't sorted.
1356 void radiusMatchCollection(const oclMat &query, oclMat &trainIdx, oclMat &imgIdx, oclMat &distance, oclMat &nMatches, float maxDistance,
1357 const std::vector<oclMat> &masks = std::vector<oclMat>());
1359 // Download trainIdx, imgIdx, nMatches and distance and convert it to vector with DMatch.
1360 // matches will be sorted in increasing order of distances.
1361 // compactResult is used when mask is not empty. If compactResult is false matches
1362 // vector will have the same size as queryDescriptors rows. If compactResult is true
1363 // matches vector will not contain matches for fully masked out query descriptors.
1364 static void radiusMatchDownload(const oclMat &trainIdx, const oclMat &imgIdx, const oclMat &distance, const oclMat &nMatches,
1365 std::vector< std::vector<DMatch> > &matches, bool compactResult = false);
1366 // Convert trainIdx, nMatches and distance to vector with DMatch.
1367 static void radiusMatchConvert(const Mat &trainIdx, const Mat &imgIdx, const Mat &distance, const Mat &nMatches,
1368 std::vector< std::vector<DMatch> > &matches, bool compactResult = false);
1370 // Find best matches from train collection for each query descriptor which have distance less than
1371 // maxDistance (in increasing order of distances).
1372 void radiusMatch(const oclMat &query, std::vector< std::vector<DMatch> > &matches, float maxDistance,
1373 const std::vector<oclMat> &masks = std::vector<oclMat>(), bool compactResult = false);
1378 std::vector<oclMat> trainDescCollection;
1381 template <class Distance>
1382 class CV_EXPORTS BruteForceMatcher_OCL;
1384 template <typename T>
1385 class CV_EXPORTS BruteForceMatcher_OCL< L1<T> > : public BruteForceMatcher_OCL_base
1388 explicit BruteForceMatcher_OCL() : BruteForceMatcher_OCL_base(L1Dist) {}
1389 explicit BruteForceMatcher_OCL(L1<T> /*d*/) : BruteForceMatcher_OCL_base(L1Dist) {}
1391 template <typename T>
1392 class CV_EXPORTS BruteForceMatcher_OCL< L2<T> > : public BruteForceMatcher_OCL_base
1395 explicit BruteForceMatcher_OCL() : BruteForceMatcher_OCL_base(L2Dist) {}
1396 explicit BruteForceMatcher_OCL(L2<T> /*d*/) : BruteForceMatcher_OCL_base(L2Dist) {}
1398 template <> class CV_EXPORTS BruteForceMatcher_OCL< Hamming > : public BruteForceMatcher_OCL_base
1401 explicit BruteForceMatcher_OCL() : BruteForceMatcher_OCL_base(HammingDist) {}
1402 explicit BruteForceMatcher_OCL(Hamming /*d*/) : BruteForceMatcher_OCL_base(HammingDist) {}
1405 class CV_EXPORTS BFMatcher_OCL : public BruteForceMatcher_OCL_base
1408 explicit BFMatcher_OCL(int norm = NORM_L2) : BruteForceMatcher_OCL_base(norm == NORM_L1 ? L1Dist : norm == NORM_L2 ? L2Dist : HammingDist) {}
1411 class CV_EXPORTS GoodFeaturesToTrackDetector_OCL
1414 explicit GoodFeaturesToTrackDetector_OCL(int maxCorners = 1000, double qualityLevel = 0.01, double minDistance = 0.0,
1415 int blockSize = 3, bool useHarrisDetector = false, double harrisK = 0.04);
1417 //! return 1 rows matrix with CV_32FC2 type
1418 void operator ()(const oclMat& image, oclMat& corners, const oclMat& mask = oclMat());
1419 //! download points of type Point2f to a vector. the vector's content will be erased
1420 void downloadPoints(const oclMat &points, std::vector<Point2f> &points_v);
1423 double qualityLevel;
1427 bool useHarrisDetector;
1429 void releaseMemory()
1434 minMaxbuf_.release();
1435 tmpCorners_.release();
1445 inline GoodFeaturesToTrackDetector_OCL::GoodFeaturesToTrackDetector_OCL(int maxCorners_, double qualityLevel_, double minDistance_,
1446 int blockSize_, bool useHarrisDetector_, double harrisK_)
1448 maxCorners = maxCorners_;
1449 qualityLevel = qualityLevel_;
1450 minDistance = minDistance_;
1451 blockSize = blockSize_;
1452 useHarrisDetector = useHarrisDetector_;
1456 /////////////////////////////// PyrLKOpticalFlow /////////////////////////////////////
1458 class CV_EXPORTS PyrLKOpticalFlow
1463 winSize = Size(21, 21);
1467 useInitialFlow = false;
1468 minEigThreshold = 1e-4f;
1469 getMinEigenVals = false;
1470 isDeviceArch11_ = false;
1473 void sparse(const oclMat &prevImg, const oclMat &nextImg, const oclMat &prevPts, oclMat &nextPts,
1474 oclMat &status, oclMat *err = 0);
1476 void dense(const oclMat &prevImg, const oclMat &nextImg, oclMat &u, oclMat &v, oclMat *err = 0);
1482 bool useInitialFlow;
1483 float minEigThreshold;
1484 bool getMinEigenVals;
1486 void releaseMemory()
1488 dx_calcBuf_.release();
1489 dy_calcBuf_.release();
1499 void calcSharrDeriv(const oclMat &src, oclMat &dx, oclMat &dy);
1501 void buildImagePyramid(const oclMat &img0, std::vector<oclMat> &pyr, bool withBorder);
1506 std::vector<oclMat> prevPyr_;
1507 std::vector<oclMat> nextPyr_;
1515 bool isDeviceArch11_;
1518 class CV_EXPORTS FarnebackOpticalFlow
1521 FarnebackOpticalFlow();
1532 void operator ()(const oclMat &frame0, const oclMat &frame1, oclMat &flowx, oclMat &flowy);
1534 void releaseMemory();
1537 void prepareGaussian(
1538 int n, double sigma, float *g, float *xg, float *xxg,
1539 double &ig11, double &ig03, double &ig33, double &ig55);
1541 void setPolynomialExpansionConsts(int n, double sigma);
1543 void updateFlow_boxFilter(
1544 const oclMat& R0, const oclMat& R1, oclMat& flowx, oclMat &flowy,
1545 oclMat& M, oclMat &bufM, int blockSize, bool updateMatrices);
1547 void updateFlow_gaussianBlur(
1548 const oclMat& R0, const oclMat& R1, oclMat& flowx, oclMat& flowy,
1549 oclMat& M, oclMat &bufM, int blockSize, bool updateMatrices);
1552 oclMat pyrLevel_[2], M_, bufM_, R_[2], blurredFrame_[2];
1553 std::vector<oclMat> pyramid0_, pyramid1_;
1556 //////////////// build warping maps ////////////////////
1557 //! builds plane warping maps
1558 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);
1559 //! builds cylindrical warping maps
1560 CV_EXPORTS void buildWarpCylindricalMaps(Size src_size, Rect dst_roi, const Mat &K, const Mat &R, float scale, oclMat &map_x, oclMat &map_y);
1561 //! builds spherical warping maps
1562 CV_EXPORTS void buildWarpSphericalMaps(Size src_size, Rect dst_roi, const Mat &K, const Mat &R, float scale, oclMat &map_x, oclMat &map_y);
1563 //! builds Affine warping maps
1564 CV_EXPORTS void buildWarpAffineMaps(const Mat &M, bool inverse, Size dsize, oclMat &xmap, oclMat &ymap);
1566 //! builds Perspective warping maps
1567 CV_EXPORTS void buildWarpPerspectiveMaps(const Mat &M, bool inverse, Size dsize, oclMat &xmap, oclMat &ymap);
1569 ///////////////////////////////////// interpolate frames //////////////////////////////////////////////
1570 //! Interpolate frames (images) using provided optical flow (displacement field).
1571 //! frame0 - frame 0 (32-bit floating point images, single channel)
1572 //! frame1 - frame 1 (the same type and size)
1573 //! fu - forward horizontal displacement
1574 //! fv - forward vertical displacement
1575 //! bu - backward horizontal displacement
1576 //! bv - backward vertical displacement
1577 //! pos - new frame position
1578 //! newFrame - new frame
1579 //! buf - temporary buffer, will have width x 6*height size, CV_32FC1 type and contain 6 oclMat;
1580 //! occlusion masks 0, occlusion masks 1,
1581 //! interpolated forward flow 0, interpolated forward flow 1,
1582 //! interpolated backward flow 0, interpolated backward flow 1
1584 CV_EXPORTS void interpolateFrames(const oclMat &frame0, const oclMat &frame1,
1585 const oclMat &fu, const oclMat &fv,
1586 const oclMat &bu, const oclMat &bv,
1587 float pos, oclMat &newFrame, oclMat &buf);
1589 //! computes moments of the rasterized shape or a vector of points
1590 CV_EXPORTS Moments ocl_moments(InputArray _array, bool binaryImage);
1592 class CV_EXPORTS StereoBM_OCL
1595 enum { BASIC_PRESET = 0, PREFILTER_XSOBEL = 1 };
1597 enum { DEFAULT_NDISP = 64, DEFAULT_WINSZ = 19 };
1599 //! the default constructor
1601 //! the full constructor taking the camera-specific preset, number of disparities and the SAD window size. ndisparities must be multiple of 8.
1602 StereoBM_OCL(int preset, int ndisparities = DEFAULT_NDISP, int winSize = DEFAULT_WINSZ);
1604 //! the stereo correspondence operator. Finds the disparity for the specified rectified stereo pair
1605 //! Output disparity has CV_8U type.
1606 void operator() ( const oclMat &left, const oclMat &right, oclMat &disparity);
1608 //! Some heuristics that tries to estmate
1609 // if current GPU will be faster then CPU in this algorithm.
1610 // It queries current active device.
1611 static bool checkIfGpuCallReasonable();
1617 // If avergeTexThreshold == 0 => post procesing is disabled
1618 // If avergeTexThreshold != 0 then disparity is set 0 in each point (x,y) where for left image
1619 // SumOfHorizontalGradiensInWindow(x, y, winSize) < (winSize * winSize) * avergeTexThreshold
1620 // i.e. input left image is low textured.
1621 float avergeTexThreshold;
1623 oclMat minSSD, leBuf, riBuf;
1626 class CV_EXPORTS StereoBeliefPropagation
1629 enum { DEFAULT_NDISP = 64 };
1630 enum { DEFAULT_ITERS = 5 };
1631 enum { DEFAULT_LEVELS = 5 };
1632 static void estimateRecommendedParams(int width, int height, int &ndisp, int &iters, int &levels);
1633 explicit StereoBeliefPropagation(int ndisp = DEFAULT_NDISP,
1634 int iters = DEFAULT_ITERS,
1635 int levels = DEFAULT_LEVELS,
1636 int msg_type = CV_16S);
1637 StereoBeliefPropagation(int ndisp, int iters, int levels,
1638 float max_data_term, float data_weight,
1639 float max_disc_term, float disc_single_jump,
1640 int msg_type = CV_32F);
1641 void operator()(const oclMat &left, const oclMat &right, oclMat &disparity);
1642 void operator()(const oclMat &data, oclMat &disparity);
1646 float max_data_term;
1648 float max_disc_term;
1649 float disc_single_jump;
1652 oclMat u, d, l, r, u2, d2, l2, r2;
1653 std::vector<oclMat> datas;
1657 class CV_EXPORTS StereoConstantSpaceBP
1660 enum { DEFAULT_NDISP = 128 };
1661 enum { DEFAULT_ITERS = 8 };
1662 enum { DEFAULT_LEVELS = 4 };
1663 enum { DEFAULT_NR_PLANE = 4 };
1664 static void estimateRecommendedParams(int width, int height, int &ndisp, int &iters, int &levels, int &nr_plane);
1665 explicit StereoConstantSpaceBP(
1666 int ndisp = DEFAULT_NDISP,
1667 int iters = DEFAULT_ITERS,
1668 int levels = DEFAULT_LEVELS,
1669 int nr_plane = DEFAULT_NR_PLANE,
1670 int msg_type = CV_32F);
1671 StereoConstantSpaceBP(int ndisp, int iters, int levels, int nr_plane,
1672 float max_data_term, float data_weight, float max_disc_term, float disc_single_jump,
1673 int min_disp_th = 0,
1674 int msg_type = CV_32F);
1675 void operator()(const oclMat &left, const oclMat &right, oclMat &disparity);
1680 float max_data_term;
1682 float max_disc_term;
1683 float disc_single_jump;
1686 bool use_local_init_data_cost;
1688 oclMat u[2], d[2], l[2], r[2];
1689 oclMat disp_selected_pyr[2];
1691 oclMat data_cost_selected;
1696 // Implementation of the Zach, Pock and Bischof Dual TV-L1 Optical Flow method
1699 // [1] C. Zach, T. Pock and H. Bischof, "A Duality Based Approach for Realtime TV-L1 Optical Flow".
1700 // [2] Javier Sanchez, Enric Meinhardt-Llopis and Gabriele Facciolo. "TV-L1 Optical Flow Estimation".
1701 class CV_EXPORTS OpticalFlowDual_TVL1_OCL
1704 OpticalFlowDual_TVL1_OCL();
1706 void operator ()(const oclMat& I0, const oclMat& I1, oclMat& flowx, oclMat& flowy);
1708 void collectGarbage();
1711 * Time step of the numerical scheme.
1716 * Weight parameter for the data term, attachment parameter.
1717 * This is the most relevant parameter, which determines the smoothness of the output.
1718 * The smaller this parameter is, the smoother the solutions we obtain.
1719 * It depends on the range of motions of the images, so its value should be adapted to each image sequence.
1724 * Weight parameter for (u - v)^2, tightness parameter.
1725 * It serves as a link between the attachment and the regularization terms.
1726 * In theory, it should have a small value in order to maintain both parts in correspondence.
1727 * The method is stable for a large range of values of this parameter.
1732 * Number of scales used to create the pyramid of images.
1737 * Number of warpings per scale.
1738 * Represents the number of times that I1(x+u0) and grad( I1(x+u0) ) are computed per scale.
1739 * This is a parameter that assures the stability of the method.
1740 * It also affects the running time, so it is a compromise between speed and accuracy.
1745 * Stopping criterion threshold used in the numerical scheme, which is a trade-off between precision and running time.
1746 * A small value will yield more accurate solutions at the expense of a slower convergence.
1751 * Stopping criterion iterations number used in the numerical scheme.
1755 bool useInitialFlow;
1758 void procOneScale(const oclMat& I0, const oclMat& I1, oclMat& u1, oclMat& u2);
1760 std::vector<oclMat> I0s;
1761 std::vector<oclMat> I1s;
1762 std::vector<oclMat> u1s;
1763 std::vector<oclMat> u2s;
1783 // current supported sorting methods
1786 SORT_BITONIC, // only support power-of-2 buffer size
1787 SORT_SELECTION, // cannot sort duplicate keys
1789 SORT_RADIX // only support signed int/float keys(CV_32S/CV_32F)
1791 //! Returns the sorted result of all the elements in input based on equivalent keys.
1793 // The element unit in the values to be sorted is determined from the data type,
1794 // i.e., a CV_32FC2 input {a1a2, b1b2} will be considered as two elements, regardless its
1795 // matrix dimension.
1796 // both keys and values will be sorted inplace
1797 // Key needs to be single channel oclMat.
1801 // keys = {2, 3, 1} (CV_8UC1)
1802 // values = {10,5, 4,3, 6,2} (CV_8UC2)
1803 // sortByKey(keys, values, SORT_SELECTION, false);
1805 // keys = {1, 2, 3} (CV_8UC1)
1806 // values = {6,2, 10,5, 4,3} (CV_8UC2)
1807 void CV_EXPORTS sortByKey(oclMat& keys, oclMat& values, int method, bool isGreaterThan = false);
1808 /*!Base class for MOG and MOG2!*/
1809 class CV_EXPORTS BackgroundSubtractor
1812 //! the virtual destructor
1813 virtual ~BackgroundSubtractor();
1814 //! the update operator that takes the next video frame and returns the current foreground mask as 8-bit binary image.
1815 virtual void operator()(const oclMat& image, oclMat& fgmask, float learningRate);
1817 //! computes a background image
1818 virtual void getBackgroundImage(oclMat& backgroundImage) const = 0;
1821 Gaussian Mixture-based Backbround/Foreground Segmentation Algorithm
1823 The class implements the following algorithm:
1824 "An improved adaptive background mixture model for real-time tracking with shadow detection"
1825 P. KadewTraKuPong and R. Bowden,
1826 Proc. 2nd European Workshp on Advanced Video-Based Surveillance Systems, 2001."
1827 http://personal.ee.surrey.ac.uk/Personal/R.Bowden/publications/avbs01/avbs01.pdf
1829 class CV_EXPORTS MOG: public cv::ocl::BackgroundSubtractor
1832 //! the default constructor
1833 MOG(int nmixtures = -1);
1835 //! re-initiaization method
1836 void initialize(Size frameSize, int frameType);
1838 //! the update operator
1839 void operator()(const oclMat& frame, oclMat& fgmask, float learningRate = 0.f);
1841 //! computes a background image which are the mean of all background gaussians
1842 void getBackgroundImage(oclMat& backgroundImage) const;
1844 //! releases all inner buffers
1849 float backgroundRatio;
1866 The class implements the following algorithm:
1867 "Improved adaptive Gausian mixture model for background subtraction"
1869 International Conference Pattern Recognition, UK, August, 2004.
1870 http://www.zoranz.net/Publications/zivkovic2004ICPR.pdf
1872 class CV_EXPORTS MOG2: public cv::ocl::BackgroundSubtractor
1875 //! the default constructor
1876 MOG2(int nmixtures = -1);
1878 //! re-initiaization method
1879 void initialize(Size frameSize, int frameType);
1881 //! the update operator
1882 void operator()(const oclMat& frame, oclMat& fgmask, float learningRate = -1.0f);
1884 //! computes a background image which are the mean of all background gaussians
1885 void getBackgroundImage(oclMat& backgroundImage) const;
1887 //! releases all inner buffers
1891 // you should call initialize after parameters changes
1895 //! here it is the maximum allowed number of mixture components.
1896 //! Actual number is determined dynamically per pixel
1898 // threshold on the squared Mahalanobis distance to decide if it is well described
1899 // by the background model or not. Related to Cthr from the paper.
1900 // This does not influence the update of the background. A typical value could be 4 sigma
1901 // and that is varThreshold=4*4=16; Corresponds to Tb in the paper.
1903 /////////////////////////
1904 // less important parameters - things you might change but be carefull
1905 ////////////////////////
1907 float backgroundRatio;
1908 // corresponds to fTB=1-cf from the paper
1909 // TB - threshold when the component becomes significant enough to be included into
1910 // the background model. It is the TB=1-cf from the paper. So I use cf=0.1 => TB=0.
1911 // For alpha=0.001 it means that the mode should exist for approximately 105 frames before
1912 // it is considered foreground
1913 // float noiseSigma;
1914 float varThresholdGen;
1916 //correspondts to Tg - threshold on the squared Mahalan. dist. to decide
1917 //when a sample is close to the existing components. If it is not close
1918 //to any a new component will be generated. I use 3 sigma => Tg=3*3=9.
1919 //Smaller Tg leads to more generated components and higher Tg might make
1920 //lead to small number of components but they can grow too large
1925 //initial variance for the newly generated components.
1926 //It will will influence the speed of adaptation. A good guess should be made.
1927 //A simple way is to estimate the typical standard deviation from the images.
1928 //I used here 10 as a reasonable value
1929 // min and max can be used to further control the variance
1930 float fCT; //CT - complexity reduction prior
1931 //this is related to the number of samples needed to accept that a component
1932 //actually exists. We use CT=0.05 of all the samples. By setting CT=0 you get
1933 //the standard Stauffer&Grimson algorithm (maybe not exact but very similar)
1935 //shadow detection parameters
1936 bool bShadowDetection; //default 1 - do shadow detection
1937 unsigned char nShadowDetection; //do shadow detection - insert this value as the detection result - 127 default value
1939 // Tau - shadow threshold. The shadow is detected if the pixel is darker
1940 //version of the background. Tau is a threshold on how much darker the shadow can be.
1941 //Tau= 0.5 means that if pixel is more than 2 times darker then it is not shadow
1942 //See: Prati,Mikic,Trivedi,Cucchiarra,"Detecting Moving Shadows...",IEEE PAMI,2003.
1955 oclMat bgmodelUsedModes_; //keep track of number of modes per pixel
1959 #if defined _MSC_VER && _MSC_VER >= 1200
1960 # pragma warning( push)
1961 # pragma warning( disable: 4267)
1963 #include "opencv2/ocl/matrix_operations.hpp"
1964 #if defined _MSC_VER && _MSC_VER >= 1200
1965 # pragma warning( pop)
1968 #endif /* __OPENCV_OCL_HPP__ */