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/core.hpp"
51 #include "opencv2/imgproc/imgproc.hpp"
52 #include "opencv2/objdetect/objdetect.hpp"
53 #include "opencv2/features2d/features2d.hpp"
54 #include "opencv2/ml/ml.hpp"
62 CVCL_DEVICE_TYPE_DEFAULT = (1 << 0),
63 CVCL_DEVICE_TYPE_CPU = (1 << 1),
64 CVCL_DEVICE_TYPE_GPU = (1 << 2),
65 CVCL_DEVICE_TYPE_ACCELERATOR = (1 << 3),
66 //CVCL_DEVICE_TYPE_CUSTOM = (1 << 4)
67 CVCL_DEVICE_TYPE_ALL = 0xFFFFFFFF
79 DEVICE_MEM_DEFAULT = 0,
80 DEVICE_MEM_AHP, //alloc host pointer
81 DEVICE_MEM_UHP, //use host pointer
82 DEVICE_MEM_CHP, //copy host pointer
83 DEVICE_MEM_PM //persistent memory
86 // 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
115 std::string compilationExtraOptions;
117 const PlatformInfo* platform;
124 int _id; // reserved, don't use it
126 std::string platformProfile;
127 std::string platformVersion;
128 std::string platformName;
129 std::string platformVendor;
130 std::string platformExtensons;
132 int platformVersionMajor;
133 int platformVersionMinor;
135 std::vector<const DeviceInfo*> devices;
140 //////////////////////////////// Initialization & Info ////////////////////////
141 typedef std::vector<const PlatformInfo*> PlatformsInfo;
143 CV_EXPORTS int getOpenCLPlatforms(PlatformsInfo& platforms);
145 typedef std::vector<const DeviceInfo*> DevicesInfo;
147 CV_EXPORTS int getOpenCLDevices(DevicesInfo& devices, int deviceType = CVCL_DEVICE_TYPE_GPU,
148 const PlatformInfo* platform = NULL);
150 // set device you want to use
151 CV_EXPORTS void setDevice(const DeviceInfo* info);
153 //////////////////////////////// Error handling ////////////////////////
154 CV_EXPORTS void error(const char *error_string, const char *file, const int line, const char *func);
158 FEATURE_CL_DOUBLE = 1,
159 FEATURE_CL_UNIFIED_MEM,
163 // Represents OpenCL context, interface
164 class CV_EXPORTS Context
170 static Context* getContext();
172 bool supportsFeature(FEATURE_TYPE featureType) const;
173 const DeviceInfo& getDeviceInfo() const;
175 const void* getOpenCLContextPtr() const;
176 const void* getOpenCLCommandQueuePtr() const;
177 const void* getOpenCLDeviceIDPtr() const;
180 inline const void *getClContextPtr()
182 return Context::getContext()->getOpenCLContextPtr();
185 inline const void *getClCommandQueuePtr()
187 return Context::getContext()->getOpenCLCommandQueuePtr();
190 CV_EXPORTS bool supportsFeature(FEATURE_TYPE featureType);
192 CV_EXPORTS void finish();
194 enum BINARY_CACHE_MODE
196 CACHE_NONE = 0, // do not cache OpenCL binary
197 CACHE_DEBUG = 0x1 << 0, // cache OpenCL binary when built in debug mode
198 CACHE_RELEASE = 0x1 << 1, // default behavior, only cache when built in release mode
199 CACHE_ALL = CACHE_DEBUG | CACHE_RELEASE, // cache opencl binary
201 //! Enable or disable OpenCL program binary caching onto local disk
202 // After a program (*.cl files in opencl/ folder) is built at runtime, we allow the
203 // compiled OpenCL program to be cached to the path automatically as "path/*.clb"
204 // binary file, which will be reused when the OpenCV executable is started again.
206 // This feature is enabled by default.
207 CV_EXPORTS void setBinaryDiskCache(int mode = CACHE_RELEASE, cv::String path = "./");
209 //! set where binary cache to be saved to
210 CV_EXPORTS void setBinaryPath(const char *path);
215 const char* programStr;
216 const char* programHash;
218 // Cache in memory by name (should be unique). Caching on disk disabled.
219 inline ProgramSource(const char* _name, const char* _programStr)
220 : name(_name), programStr(_programStr), programHash(NULL)
224 // Cache in memory by name (should be unique). Caching on disk uses programHash mark.
225 inline ProgramSource(const char* _name, const char* _programStr, const char* _programHash)
226 : name(_name), programStr(_programStr), programHash(_programHash)
231 //! Calls OpenCL kernel. Pass globalThreads = NULL, and cleanUp = true, to finally clean-up without executing.
232 //! Deprecated, will be replaced
233 CV_EXPORTS void openCLExecuteKernelInterop(Context *clCxt,
234 const cv::ocl::ProgramSource& source, string kernelName,
235 size_t globalThreads[3], size_t localThreads[3],
236 std::vector< std::pair<size_t, const void *> > &args,
237 int channels, int depth, const char *build_options);
239 class CV_EXPORTS oclMatExpr;
240 //////////////////////////////// oclMat ////////////////////////////////
241 class CV_EXPORTS oclMat
244 //! default constructor
246 //! constructs oclMatrix of the specified size and type (_type is CV_8UC1, CV_64FC3, CV_32SC(12) etc.)
247 oclMat(int rows, int cols, int type);
248 oclMat(Size size, int type);
249 //! constucts oclMatrix and fills it with the specified value _s.
250 oclMat(int rows, int cols, int type, const Scalar &s);
251 oclMat(Size size, int type, const Scalar &s);
253 oclMat(const oclMat &m);
255 //! constructor for oclMatrix headers pointing to user-allocated data
256 oclMat(int rows, int cols, int type, void *data, size_t step = Mat::AUTO_STEP);
257 oclMat(Size size, int type, void *data, size_t step = Mat::AUTO_STEP);
259 //! creates a matrix header for a part of the bigger matrix
260 oclMat(const oclMat &m, const Range &rowRange, const Range &colRange);
261 oclMat(const oclMat &m, const Rect &roi);
263 //! builds oclMat from Mat. Perfom blocking upload to device.
264 explicit oclMat (const Mat &m);
266 //! destructor - calls release()
269 //! assignment operators
270 oclMat &operator = (const oclMat &m);
271 //! assignment operator. Perfom blocking upload to device.
272 oclMat &operator = (const Mat &m);
273 oclMat &operator = (const oclMatExpr& expr);
275 //! pefroms blocking upload data to oclMat.
276 void upload(const cv::Mat &m);
279 //! downloads data from device to host memory. Blocking calls.
280 operator Mat() const;
281 void download(cv::Mat &m) const;
283 //! convert to _InputArray
284 operator _InputArray();
286 //! convert to _OutputArray
287 operator _OutputArray();
289 //! returns a new oclMatrix header for the specified row
290 oclMat row(int y) const;
291 //! returns a new oclMatrix header for the specified column
292 oclMat col(int x) const;
293 //! ... for the specified row span
294 oclMat rowRange(int startrow, int endrow) const;
295 oclMat rowRange(const Range &r) const;
296 //! ... for the specified column span
297 oclMat colRange(int startcol, int endcol) const;
298 oclMat colRange(const Range &r) const;
300 //! returns deep copy of the oclMatrix, i.e. the data is copied
301 oclMat clone() const;
303 //! copies those oclMatrix elements to "m" that are marked with non-zero mask elements.
304 // It calls m.create(this->size(), this->type()).
305 // It supports any data type
306 void copyTo( oclMat &m, const oclMat &mask = oclMat()) const;
308 //! converts oclMatrix to another datatype with optional scalng. See cvConvertScale.
309 //It supports 8UC1 8UC4 32SC1 32SC4 32FC1 32FC4
310 void convertTo( oclMat &m, int rtype, double alpha = 1, double beta = 0 ) const;
312 void assignTo( oclMat &m, int type = -1 ) const;
314 //! sets every oclMatrix element to s
315 //It supports 8UC1 8UC4 32SC1 32SC4 32FC1 32FC4
316 oclMat& operator = (const Scalar &s);
317 //! sets some of the oclMatrix elements to s, according to the mask
318 //It supports 8UC1 8UC4 32SC1 32SC4 32FC1 32FC4
319 oclMat& setTo(const Scalar &s, const oclMat &mask = oclMat());
320 //! creates alternative oclMatrix header for the same data, with different
321 // number of channels and/or different number of rows. see cvReshape.
322 oclMat reshape(int cn, int rows = 0) const;
324 //! allocates new oclMatrix data unless the oclMatrix already has specified size and type.
325 // previous data is unreferenced if needed.
326 void create(int rows, int cols, int type);
327 void create(Size size, int type);
329 //! allocates new oclMatrix with specified device memory type.
330 void createEx(int rows, int cols, int type, DevMemRW rw_type, DevMemType mem_type);
331 void createEx(Size size, int type, DevMemRW rw_type, DevMemType mem_type);
333 //! decreases reference counter;
334 // deallocate the data when reference counter reaches 0.
337 //! swaps with other smart pointer
338 void swap(oclMat &mat);
340 //! locates oclMatrix header within a parent oclMatrix. See below
341 void locateROI( Size &wholeSize, Point &ofs ) const;
342 //! moves/resizes the current oclMatrix ROI inside the parent oclMatrix.
343 oclMat& adjustROI( int dtop, int dbottom, int dleft, int dright );
344 //! extracts a rectangular sub-oclMatrix
345 // (this is a generalized form of row, rowRange etc.)
346 oclMat operator()( Range rowRange, Range colRange ) const;
347 oclMat operator()( const Rect &roi ) const;
349 oclMat& operator+=( const oclMat& m );
350 oclMat& operator-=( const oclMat& m );
351 oclMat& operator*=( const oclMat& m );
352 oclMat& operator/=( const oclMat& m );
354 //! returns true if the oclMatrix data is continuous
355 // (i.e. when there are no gaps between successive rows).
356 // similar to CV_IS_oclMat_CONT(cvoclMat->type)
357 bool isContinuous() const;
358 //! returns element size in bytes,
359 // similar to CV_ELEM_SIZE(cvMat->type)
360 size_t elemSize() const;
361 //! returns the size of element channel in bytes.
362 size_t elemSize1() const;
363 //! returns element type, similar to CV_MAT_TYPE(cvMat->type)
365 //! returns element type, i.e. 8UC3 returns 8UC4 because in ocl
366 //! 3 channels element actually use 4 channel space
368 //! returns element type, similar to CV_MAT_DEPTH(cvMat->type)
370 //! returns element type, similar to CV_MAT_CN(cvMat->type)
371 int channels() const;
372 //! returns element type, return 4 for 3 channels element,
373 //!becuase 3 channels element actually use 4 channel space
374 int oclchannels() const;
375 //! returns step/elemSize1()
376 size_t step1() const;
377 //! returns oclMatrix size:
378 // width == number of columns, height == number of rows
380 //! returns true if oclMatrix data is NULL
383 //! returns pointer to y-th row
384 uchar* ptr(int y = 0);
385 const uchar *ptr(int y = 0) const;
387 //! template version of the above method
388 template<typename _Tp> _Tp *ptr(int y = 0);
389 template<typename _Tp> const _Tp *ptr(int y = 0) const;
391 //! matrix transposition
394 /*! includes several bit-fields:
395 - the magic signature
401 //! the number of rows and columns
403 //! a distance between successive rows in bytes; includes the gap if any
405 //! pointer to the data(OCL memory object)
408 //! pointer to the reference counter;
409 // when oclMatrix points to user-allocated data, the pointer is NULL
412 //! helper fields used in locateROI and adjustROI
413 //datastart and dataend are not used in current version
417 //! OpenCL context associated with the oclMat object.
418 Context *clCxt; // TODO clCtx
419 //add offset for handle ROI, calculated in byte
421 //add wholerows and wholecols for the whole matrix, datastart and dataend are no longer used
426 // convert InputArray/OutputArray to oclMat references
427 CV_EXPORTS oclMat& getOclMatRef(InputArray src);
428 CV_EXPORTS oclMat& getOclMatRef(OutputArray src);
430 ///////////////////// mat split and merge /////////////////////////////////
431 //! Compose a multi-channel array from several single-channel arrays
433 CV_EXPORTS void merge(const oclMat *src, size_t n, oclMat &dst);
434 CV_EXPORTS void merge(const vector<oclMat> &src, oclMat &dst);
436 //! Divides multi-channel array into several single-channel arrays
438 CV_EXPORTS void split(const oclMat &src, oclMat *dst);
439 CV_EXPORTS void split(const oclMat &src, vector<oclMat> &dst);
441 ////////////////////////////// Arithmetics ///////////////////////////////////
443 //! adds one matrix to another with scale (dst = src1 * alpha + src2 * beta + gama)
444 // supports all data types
445 CV_EXPORTS void addWeighted(const oclMat &src1, double alpha, const oclMat &src2, double beta, double gama, oclMat &dst);
447 //! adds one matrix to another (dst = src1 + src2)
448 // supports all data types
449 CV_EXPORTS void add(const oclMat &src1, const oclMat &src2, oclMat &dst, const oclMat &mask = oclMat());
450 //! adds scalar to a matrix (dst = src1 + s)
451 // supports all data types
452 CV_EXPORTS void add(const oclMat &src1, const Scalar &s, oclMat &dst, const oclMat &mask = oclMat());
454 //! subtracts one matrix from another (dst = src1 - src2)
455 // supports all data types
456 CV_EXPORTS void subtract(const oclMat &src1, const oclMat &src2, oclMat &dst, const oclMat &mask = oclMat());
457 //! subtracts scalar from a matrix (dst = src1 - s)
458 // supports all data types
459 CV_EXPORTS void subtract(const oclMat &src1, const Scalar &s, oclMat &dst, const oclMat &mask = oclMat());
461 //! computes element-wise product of the two arrays (dst = src1 * scale * src2)
462 // supports all data types
463 CV_EXPORTS void multiply(const oclMat &src1, const oclMat &src2, oclMat &dst, double scale = 1);
464 //! multiplies matrix to a number (dst = scalar * src)
465 // supports all data types
466 CV_EXPORTS void multiply(double scalar, const oclMat &src, oclMat &dst);
468 //! computes element-wise quotient of the two arrays (dst = src1 * scale / src2)
469 // supports all data types
470 CV_EXPORTS void divide(const oclMat &src1, const oclMat &src2, oclMat &dst, double scale = 1);
471 //! computes element-wise quotient of the two arrays (dst = scale / src)
472 // supports all data types
473 CV_EXPORTS void divide(double scale, const oclMat &src1, oclMat &dst);
475 //! computes element-wise minimum of the two arrays (dst = min(src1, src2))
476 // supports all data types
477 CV_EXPORTS void min(const oclMat &src1, const oclMat &src2, oclMat &dst);
479 //! computes element-wise maximum of the two arrays (dst = max(src1, src2))
480 // supports all data types
481 CV_EXPORTS void max(const oclMat &src1, const oclMat &src2, oclMat &dst);
483 //! compares elements of two arrays (dst = src1 <cmpop> src2)
484 // supports all data types
485 CV_EXPORTS void compare(const oclMat &src1, const oclMat &src2, oclMat &dst, int cmpop);
487 //! transposes the matrix
488 // supports all data types
489 CV_EXPORTS void transpose(const oclMat &src, oclMat &dst);
491 //! computes element-wise absolute values of an array (dst = abs(src))
492 // supports all data types
493 CV_EXPORTS void abs(const oclMat &src, oclMat &dst);
495 //! computes element-wise absolute difference of two arrays (dst = abs(src1 - src2))
496 // supports all data types
497 CV_EXPORTS void absdiff(const oclMat &src1, const oclMat &src2, oclMat &dst);
498 //! computes element-wise absolute difference of array and scalar (dst = abs(src1 - s))
499 // supports all data types
500 CV_EXPORTS void absdiff(const oclMat &src1, const Scalar &s, oclMat &dst);
502 //! computes mean value and standard deviation of all or selected array elements
503 // supports all data types
504 CV_EXPORTS void meanStdDev(const oclMat &mtx, Scalar &mean, Scalar &stddev);
506 //! computes norm of array
507 // supports NORM_INF, NORM_L1, NORM_L2
508 // supports all data types
509 CV_EXPORTS double norm(const oclMat &src1, int normType = NORM_L2);
511 //! computes norm of the difference between two arrays
512 // supports NORM_INF, NORM_L1, NORM_L2
513 // supports all data types
514 CV_EXPORTS double norm(const oclMat &src1, const oclMat &src2, int normType = NORM_L2);
516 //! reverses the order of the rows, columns or both in a matrix
517 // supports all types
518 CV_EXPORTS void flip(const oclMat &src, oclMat &dst, int flipCode);
520 //! computes sum of array elements
522 CV_EXPORTS Scalar sum(const oclMat &m);
523 CV_EXPORTS Scalar absSum(const oclMat &m);
524 CV_EXPORTS Scalar sqrSum(const oclMat &m);
526 //! finds global minimum and maximum array elements and returns their values
527 // support all C1 types
528 CV_EXPORTS void minMax(const oclMat &src, double *minVal, double *maxVal = 0, const oclMat &mask = oclMat());
530 //! finds global minimum and maximum array elements and returns their values with locations
531 // support all C1 types
532 CV_EXPORTS void minMaxLoc(const oclMat &src, double *minVal, double *maxVal = 0, Point *minLoc = 0, Point *maxLoc = 0,
533 const oclMat &mask = oclMat());
535 //! counts non-zero array elements
537 CV_EXPORTS int countNonZero(const oclMat &src);
539 //! transforms 8-bit unsigned integers using lookup table: dst(i)=lut(src(i))
540 // destination array will have the depth type as lut and the same channels number as source
541 //It supports 8UC1 8UC4 only
542 CV_EXPORTS void LUT(const oclMat &src, const oclMat &lut, oclMat &dst);
544 //! only 8UC1 and 256 bins is supported now
545 CV_EXPORTS void calcHist(const oclMat &mat_src, oclMat &mat_hist);
546 //! only 8UC1 and 256 bins is supported now
547 CV_EXPORTS void equalizeHist(const oclMat &mat_src, oclMat &mat_dst);
549 //! only 8UC1 is supported now
550 CV_EXPORTS Ptr<cv::CLAHE> createCLAHE(double clipLimit = 40.0, Size tileGridSize = Size(8, 8));
553 // supports 8UC1 8UC4
554 CV_EXPORTS void bilateralFilter(const oclMat& src, oclMat& dst, int d, double sigmaColor, double sigmaSpace, int borderType=BORDER_DEFAULT);
556 //! Applies an adaptive bilateral filter to the input image
557 // This is not truly a bilateral filter. Instead of using user provided fixed parameters,
558 // the function calculates a constant at each window based on local standard deviation,
559 // and use this constant to do filtering.
560 // supports 8UC1, 8UC3
561 CV_EXPORTS void adaptiveBilateralFilter(const oclMat& src, oclMat& dst, Size ksize, double sigmaSpace, Point anchor = Point(-1, -1), int borderType=BORDER_DEFAULT);
563 //! computes exponent of each matrix element (dst = e**src)
564 // supports only CV_32FC1, CV_64FC1 type
565 CV_EXPORTS void exp(const oclMat &src, oclMat &dst);
567 //! computes natural logarithm of absolute value of each matrix element: dst = log(abs(src))
568 // supports only CV_32FC1, CV_64FC1 type
569 CV_EXPORTS void log(const oclMat &src, oclMat &dst);
571 //! computes magnitude of each (x(i), y(i)) vector
572 // supports only CV_32F, CV_64F type
573 CV_EXPORTS void magnitude(const oclMat &x, const oclMat &y, oclMat &magnitude);
575 //! computes angle (angle(i)) of each (x(i), y(i)) vector
576 // supports only CV_32F, CV_64F type
577 CV_EXPORTS void phase(const oclMat &x, const oclMat &y, oclMat &angle, bool angleInDegrees = false);
579 //! the function raises every element of tne input array to p
580 // support only CV_32F, CV_64F type
581 CV_EXPORTS void pow(const oclMat &x, double p, oclMat &y);
583 //! converts Cartesian coordinates to polar
584 // supports only CV_32F CV_64F type
585 CV_EXPORTS void cartToPolar(const oclMat &x, const oclMat &y, oclMat &magnitude, oclMat &angle, bool angleInDegrees = false);
587 //! converts polar coordinates to Cartesian
588 // supports only CV_32F CV_64F type
589 CV_EXPORTS void polarToCart(const oclMat &magnitude, const oclMat &angle, oclMat &x, oclMat &y, bool angleInDegrees = false);
591 //! perfroms per-elements bit-wise inversion
592 // supports all types
593 CV_EXPORTS void bitwise_not(const oclMat &src, oclMat &dst);
595 //! calculates per-element bit-wise disjunction of two arrays
596 // supports all types
597 CV_EXPORTS void bitwise_or(const oclMat &src1, const oclMat &src2, oclMat &dst, const oclMat &mask = oclMat());
598 CV_EXPORTS void bitwise_or(const oclMat &src1, const Scalar &s, oclMat &dst, const oclMat &mask = oclMat());
600 //! calculates per-element bit-wise conjunction of two arrays
601 // supports all types
602 CV_EXPORTS void bitwise_and(const oclMat &src1, const oclMat &src2, oclMat &dst, const oclMat &mask = oclMat());
603 CV_EXPORTS void bitwise_and(const oclMat &src1, const Scalar &s, oclMat &dst, const oclMat &mask = oclMat());
605 //! calculates per-element bit-wise "exclusive or" operation
606 // supports all types
607 CV_EXPORTS void bitwise_xor(const oclMat &src1, const oclMat &src2, oclMat &dst, const oclMat &mask = oclMat());
608 CV_EXPORTS void bitwise_xor(const oclMat &src1, const Scalar &s, oclMat &dst, const oclMat &mask = oclMat());
610 //! Logical operators
611 CV_EXPORTS oclMat operator ~ (const oclMat &);
612 CV_EXPORTS oclMat operator | (const oclMat &, const oclMat &);
613 CV_EXPORTS oclMat operator & (const oclMat &, const oclMat &);
614 CV_EXPORTS oclMat operator ^ (const oclMat &, const oclMat &);
617 //! Mathematics operators
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);
621 CV_EXPORTS oclMatExpr operator / (const oclMat &src1, const oclMat &src2);
623 //! computes convolution of two images
624 // support only CV_32FC1 type
625 CV_EXPORTS void convolve(const oclMat &image, const oclMat &temp1, oclMat &result);
627 CV_EXPORTS void cvtColor(const oclMat &src, oclMat &dst, int code, int dcn = 0);
629 //! initializes a scaled identity matrix
630 CV_EXPORTS void setIdentity(oclMat& src, const Scalar & val = Scalar(1));
632 //////////////////////////////// Filter Engine ////////////////////////////////
635 The Base Class for 1D or Row-wise Filters
637 This is the base class for linear or non-linear filters that process 1D data.
638 In particular, such filters are used for the "horizontal" filtering parts in separable filters.
640 class CV_EXPORTS BaseRowFilter_GPU
643 BaseRowFilter_GPU(int ksize_, int anchor_, int bordertype_) : ksize(ksize_), anchor(anchor_), bordertype(bordertype_) {}
644 virtual ~BaseRowFilter_GPU() {}
645 virtual void operator()(const oclMat &src, oclMat &dst) = 0;
646 int ksize, anchor, bordertype;
650 The Base Class for Column-wise Filters
652 This is the base class for linear or non-linear filters that process columns of 2D arrays.
653 Such filters are used for the "vertical" filtering parts in separable filters.
655 class CV_EXPORTS BaseColumnFilter_GPU
658 BaseColumnFilter_GPU(int ksize_, int anchor_, int bordertype_) : ksize(ksize_), anchor(anchor_), bordertype(bordertype_) {}
659 virtual ~BaseColumnFilter_GPU() {}
660 virtual void operator()(const oclMat &src, oclMat &dst) = 0;
661 int ksize, anchor, bordertype;
665 The Base Class for Non-Separable 2D Filters.
667 This is the base class for linear or non-linear 2D filters.
669 class CV_EXPORTS BaseFilter_GPU
672 BaseFilter_GPU(const Size &ksize_, const Point &anchor_, const int &borderType_)
673 : ksize(ksize_), anchor(anchor_), borderType(borderType_) {}
674 virtual ~BaseFilter_GPU() {}
675 virtual void operator()(const oclMat &src, oclMat &dst) = 0;
682 The Base Class for Filter Engine.
684 The class can be used to apply an arbitrary filtering operation to an image.
685 It contains all the necessary intermediate buffers.
687 class CV_EXPORTS FilterEngine_GPU
690 virtual ~FilterEngine_GPU() {}
692 virtual void apply(const oclMat &src, oclMat &dst, Rect roi = Rect(0, 0, -1, -1)) = 0;
695 //! returns the non-separable filter engine with the specified filter
696 CV_EXPORTS Ptr<FilterEngine_GPU> createFilter2D_GPU(const Ptr<BaseFilter_GPU> filter2D);
698 //! returns the primitive row filter with the specified kernel
699 CV_EXPORTS Ptr<BaseRowFilter_GPU> getLinearRowFilter_GPU(int srcType, int bufType, const Mat &rowKernel,
700 int anchor = -1, int bordertype = BORDER_DEFAULT);
702 //! returns the primitive column filter with the specified kernel
703 CV_EXPORTS Ptr<BaseColumnFilter_GPU> getLinearColumnFilter_GPU(int bufType, int dstType, const Mat &columnKernel,
704 int anchor = -1, int bordertype = BORDER_DEFAULT, double delta = 0.0);
706 //! returns the separable linear filter engine
707 CV_EXPORTS Ptr<FilterEngine_GPU> createSeparableLinearFilter_GPU(int srcType, int dstType, const Mat &rowKernel,
708 const Mat &columnKernel, const Point &anchor = Point(-1, -1), double delta = 0.0, int bordertype = BORDER_DEFAULT);
710 //! returns the separable filter engine with the specified filters
711 CV_EXPORTS Ptr<FilterEngine_GPU> createSeparableFilter_GPU(const Ptr<BaseRowFilter_GPU> &rowFilter,
712 const Ptr<BaseColumnFilter_GPU> &columnFilter);
714 //! returns the Gaussian filter engine
715 CV_EXPORTS Ptr<FilterEngine_GPU> createGaussianFilter_GPU(int type, Size ksize, double sigma1, double sigma2 = 0, int bordertype = BORDER_DEFAULT);
717 //! returns filter engine for the generalized Sobel operator
718 CV_EXPORTS Ptr<FilterEngine_GPU> createDerivFilter_GPU( int srcType, int dstType, int dx, int dy, int ksize, int borderType = BORDER_DEFAULT );
720 //! applies Laplacian operator to the image
721 // supports only ksize = 1 and ksize = 3 8UC1 8UC4 32FC1 32FC4 data type
722 CV_EXPORTS void Laplacian(const oclMat &src, oclMat &dst, int ddepth, int ksize = 1, double scale = 1);
724 //! returns 2D box filter
725 // supports CV_8UC1 and CV_8UC4 source type, dst type must be the same as source type
726 CV_EXPORTS Ptr<BaseFilter_GPU> getBoxFilter_GPU(int srcType, int dstType,
727 const Size &ksize, Point anchor = Point(-1, -1), int borderType = BORDER_DEFAULT);
729 //! returns box filter engine
730 CV_EXPORTS Ptr<FilterEngine_GPU> createBoxFilter_GPU(int srcType, int dstType, const Size &ksize,
731 const Point &anchor = Point(-1, -1), int borderType = BORDER_DEFAULT);
733 //! returns 2D filter with the specified kernel
734 // supports CV_8UC1 and CV_8UC4 types
735 CV_EXPORTS Ptr<BaseFilter_GPU> getLinearFilter_GPU(int srcType, int dstType, const Mat &kernel, const Size &ksize,
736 const Point &anchor = Point(-1, -1), int borderType = BORDER_DEFAULT);
738 //! returns the non-separable linear filter engine
739 CV_EXPORTS Ptr<FilterEngine_GPU> createLinearFilter_GPU(int srcType, int dstType, const Mat &kernel,
740 const Point &anchor = Point(-1, -1), int borderType = BORDER_DEFAULT);
742 //! smooths the image using the normalized box filter
743 // supports data type: CV_8UC1, CV_8UC4, CV_32FC1 and CV_32FC4
744 // supports border type: BORDER_CONSTANT, BORDER_REPLICATE, BORDER_REFLECT,BORDER_REFLECT_101,BORDER_WRAP
745 CV_EXPORTS void boxFilter(const oclMat &src, oclMat &dst, int ddepth, Size ksize,
746 Point anchor = Point(-1, -1), int borderType = BORDER_DEFAULT);
748 //! returns 2D morphological filter
749 //! only MORPH_ERODE and MORPH_DILATE are supported
750 // supports CV_8UC1, CV_8UC4, CV_32FC1 and CV_32FC4 types
751 // kernel must have CV_8UC1 type, one rows and cols == ksize.width * ksize.height
752 CV_EXPORTS Ptr<BaseFilter_GPU> getMorphologyFilter_GPU(int op, int type, const Mat &kernel, const Size &ksize,
753 Point anchor = Point(-1, -1));
755 //! returns morphological filter engine. Only MORPH_ERODE and MORPH_DILATE are supported.
756 CV_EXPORTS Ptr<FilterEngine_GPU> createMorphologyFilter_GPU(int op, int type, const Mat &kernel,
757 const Point &anchor = Point(-1, -1), int iterations = 1);
759 //! a synonym for normalized box filter
760 // supports data type: CV_8UC1, CV_8UC4, CV_32FC1 and CV_32FC4
761 // supports border type: BORDER_CONSTANT, BORDER_REPLICATE, BORDER_REFLECT,BORDER_REFLECT_101
762 static inline void blur(const oclMat &src, oclMat &dst, Size ksize, Point anchor = Point(-1, -1),
763 int borderType = BORDER_CONSTANT)
765 boxFilter(src, dst, -1, ksize, anchor, borderType);
768 //! applies non-separable 2D linear filter to the image
769 // Note, at the moment this function only works when anchor point is in the kernel center
770 // and kernel size supported is either 3x3 or 5x5; otherwise the function will fail to output valid result
771 CV_EXPORTS void filter2D(const oclMat &src, oclMat &dst, int ddepth, const Mat &kernel,
772 Point anchor = Point(-1, -1), int borderType = BORDER_DEFAULT);
774 //! applies separable 2D linear filter to the image
775 CV_EXPORTS void sepFilter2D(const oclMat &src, oclMat &dst, int ddepth, const Mat &kernelX, const Mat &kernelY,
776 Point anchor = Point(-1, -1), double delta = 0.0, int bordertype = BORDER_DEFAULT);
778 //! applies generalized Sobel operator to the image
779 // dst.type must equalize src.type
780 // supports data type: CV_8UC1, CV_8UC4, CV_32FC1 and CV_32FC4
781 // supports border type: BORDER_CONSTANT, BORDER_REPLICATE, BORDER_REFLECT,BORDER_REFLECT_101
782 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);
784 //! applies the vertical or horizontal Scharr operator to the image
785 // dst.type must equalize src.type
786 // supports data type: CV_8UC1, CV_8UC4, CV_32FC1 and CV_32FC4
787 // supports border type: BORDER_CONSTANT, BORDER_REPLICATE, BORDER_REFLECT,BORDER_REFLECT_101
788 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);
790 //! smooths the image using Gaussian filter.
791 // dst.type must equalize src.type
792 // supports data type: CV_8UC1, CV_8UC4, CV_32FC1 and CV_32FC4
793 // supports border type: BORDER_CONSTANT, BORDER_REPLICATE, BORDER_REFLECT,BORDER_REFLECT_101
794 CV_EXPORTS void GaussianBlur(const oclMat &src, oclMat &dst, Size ksize, double sigma1, double sigma2 = 0, int bordertype = BORDER_DEFAULT);
796 //! erodes the image (applies the local minimum operator)
797 // supports data type: CV_8UC1, CV_8UC4, CV_32FC1 and CV_32FC4
798 CV_EXPORTS void erode( const oclMat &src, oclMat &dst, const Mat &kernel, Point anchor = Point(-1, -1), int iterations = 1,
800 int borderType = BORDER_CONSTANT, const Scalar &borderValue = morphologyDefaultBorderValue());
803 //! dilates the image (applies the local maximum operator)
804 // supports data type: CV_8UC1, CV_8UC4, CV_32FC1 and CV_32FC4
805 CV_EXPORTS void dilate( const oclMat &src, oclMat &dst, const Mat &kernel, Point anchor = Point(-1, -1), int iterations = 1,
807 int borderType = BORDER_CONSTANT, const Scalar &borderValue = morphologyDefaultBorderValue());
810 //! applies an advanced morphological operation to the image
811 CV_EXPORTS void morphologyEx( const oclMat &src, oclMat &dst, int op, const Mat &kernel, Point anchor = Point(-1, -1), int iterations = 1,
813 int borderType = BORDER_CONSTANT, const Scalar &borderValue = morphologyDefaultBorderValue());
816 ////////////////////////////// Image processing //////////////////////////////
817 //! Does mean shift filtering on GPU.
818 CV_EXPORTS void meanShiftFiltering(const oclMat &src, oclMat &dst, int sp, int sr,
819 TermCriteria criteria = TermCriteria(TermCriteria::MAX_ITER + TermCriteria::EPS, 5, 1));
821 //! Does mean shift procedure on GPU.
822 CV_EXPORTS void meanShiftProc(const oclMat &src, oclMat &dstr, oclMat &dstsp, int sp, int sr,
823 TermCriteria criteria = TermCriteria(TermCriteria::MAX_ITER + TermCriteria::EPS, 5, 1));
825 //! Does mean shift segmentation with elimiation of small regions.
826 CV_EXPORTS void meanShiftSegmentation(const oclMat &src, Mat &dst, int sp, int sr, int minsize,
827 TermCriteria criteria = TermCriteria(TermCriteria::MAX_ITER + TermCriteria::EPS, 5, 1));
829 //! applies fixed threshold to the image.
830 // supports CV_8UC1 and CV_32FC1 data type
831 // supports threshold type: THRESH_BINARY, THRESH_BINARY_INV, THRESH_TRUNC, THRESH_TOZERO, THRESH_TOZERO_INV
832 CV_EXPORTS double threshold(const oclMat &src, oclMat &dst, double thresh, double maxVal, int type = THRESH_TRUNC);
834 //! resizes the image
835 // Supports INTER_NEAREST, INTER_LINEAR
836 // supports CV_8UC1, CV_8UC4, CV_32FC1 and CV_32FC4 types
837 CV_EXPORTS void resize(const oclMat &src, oclMat &dst, Size dsize, double fx = 0, double fy = 0, int interpolation = INTER_LINEAR);
839 //! Applies a generic geometrical transformation to an image.
841 // Supports INTER_NEAREST, INTER_LINEAR.
842 // Map1 supports CV_16SC2, CV_32FC2 types.
843 // Src supports CV_8UC1, CV_8UC2, CV_8UC4.
844 CV_EXPORTS void remap(const oclMat &src, oclMat &dst, oclMat &map1, oclMat &map2, int interpolation, int bordertype, const Scalar &value = Scalar());
846 //! copies 2D array to a larger destination array and pads borders with user-specifiable constant
847 // supports CV_8UC1, CV_8UC4, CV_32SC1 types
848 CV_EXPORTS void copyMakeBorder(const oclMat &src, oclMat &dst, int top, int bottom, int left, int right, int boardtype, const Scalar &value = Scalar());
850 //! Smoothes image using median filter
851 // The source 1- or 4-channel image. m should be 3 or 5, the image depth should be CV_8U or CV_32F.
852 CV_EXPORTS void medianFilter(const oclMat &src, oclMat &dst, int m);
854 //! warps the image using affine transformation
855 // Supports INTER_NEAREST, INTER_LINEAR, INTER_CUBIC
856 // supports CV_8UC1, CV_8UC4, CV_32FC1 and CV_32FC4 types
857 CV_EXPORTS void warpAffine(const oclMat &src, oclMat &dst, const Mat &M, Size dsize, int flags = INTER_LINEAR);
859 //! warps the image using perspective transformation
860 // Supports INTER_NEAREST, INTER_LINEAR, INTER_CUBIC
861 // supports CV_8UC1, CV_8UC4, CV_32FC1 and CV_32FC4 types
862 CV_EXPORTS void warpPerspective(const oclMat &src, oclMat &dst, const Mat &M, Size dsize, int flags = INTER_LINEAR);
864 //! computes the integral image and integral for the squared image
865 // sum will have CV_32S type, sqsum - CV32F type
866 // supports only CV_8UC1 source type
867 CV_EXPORTS void integral(const oclMat &src, oclMat &sum, oclMat &sqsum);
868 CV_EXPORTS void integral(const oclMat &src, oclMat &sum);
869 CV_EXPORTS void cornerHarris(const oclMat &src, oclMat &dst, int blockSize, int ksize, double k, int bordertype = cv::BORDER_DEFAULT);
870 CV_EXPORTS void cornerHarris_dxdy(const oclMat &src, oclMat &dst, oclMat &Dx, oclMat &Dy,
871 int blockSize, int ksize, double k, int bordertype = cv::BORDER_DEFAULT);
872 CV_EXPORTS void cornerMinEigenVal(const oclMat &src, oclMat &dst, int blockSize, int ksize, int bordertype = cv::BORDER_DEFAULT);
873 CV_EXPORTS void cornerMinEigenVal_dxdy(const oclMat &src, oclMat &dst, oclMat &Dx, oclMat &Dy,
874 int blockSize, int ksize, int bordertype = cv::BORDER_DEFAULT);
876 /////////////////////////////////// ML ///////////////////////////////////////////
878 //! Compute closest centers for each lines in source and lable it after center's index
879 // supports CV_32FC1/CV_32FC2/CV_32FC4 data type
880 // supports NORM_L1 and NORM_L2 distType
881 // if indices is provided, only the indexed rows will be calculated and their results are in the same
883 CV_EXPORTS void distanceToCenters(oclMat &dists, oclMat &labels, const oclMat &src, const oclMat ¢ers, int distType = NORM_L2SQR, const oclMat &indices = oclMat());
885 //!Does k-means procedure on GPU
886 // supports CV_32FC1/CV_32FC2/CV_32FC4 data type
887 CV_EXPORTS double kmeans(const oclMat &src, int K, oclMat &bestLabels,
888 TermCriteria criteria, int attemps, int flags, oclMat ¢ers);
891 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
892 ///////////////////////////////////////////CascadeClassifier//////////////////////////////////////////////////////////////////
893 ///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
895 class CV_EXPORTS_W OclCascadeClassifier : public cv::CascadeClassifier
898 OclCascadeClassifier() {};
899 ~OclCascadeClassifier() {};
901 CvSeq* oclHaarDetectObjects(oclMat &gimg, CvMemStorage *storage, double scaleFactor,
902 int minNeighbors, int flags, CvSize minSize = cvSize(0, 0), CvSize maxSize = cvSize(0, 0));
905 class CV_EXPORTS OclCascadeClassifierBuf : public cv::CascadeClassifier
908 OclCascadeClassifierBuf() :
909 m_flags(0), initialized(false), m_scaleFactor(0), buffers(NULL) {}
911 ~OclCascadeClassifierBuf() { release(); }
913 void detectMultiScale(oclMat &image, CV_OUT std::vector<cv::Rect>& faces,
914 double scaleFactor = 1.1, int minNeighbors = 3, int flags = 0,
915 Size minSize = Size(), Size maxSize = Size());
919 void Init(const int rows, const int cols, double scaleFactor, int flags,
920 const int outputsz, const size_t localThreads[],
921 CvSize minSize, CvSize maxSize);
922 void CreateBaseBufs(const int datasize, const int totalclassifier, const int flags, const int outputsz);
923 void CreateFactorRelatedBufs(const int rows, const int cols, const int flags,
924 const double scaleFactor, const size_t localThreads[],
925 CvSize minSize, CvSize maxSize);
926 void GenResult(CV_OUT std::vector<cv::Rect>& faces, const std::vector<cv::Rect> &rectList, const std::vector<int> &rweights);
933 bool findBiggestObject;
935 double m_scaleFactor;
938 vector<CvSize> sizev;
939 vector<float> scalev;
940 oclMat gimg1, gsum, gsqsum;
945 /////////////////////////////// Pyramid /////////////////////////////////////
946 CV_EXPORTS void pyrDown(const oclMat &src, oclMat &dst);
948 //! upsamples the source image and then smoothes it
949 CV_EXPORTS void pyrUp(const oclMat &src, oclMat &dst);
951 //! performs linear blending of two images
952 //! to avoid accuracy errors sum of weigths shouldn't be very close to zero
953 // supports only CV_8UC1 source type
954 CV_EXPORTS void blendLinear(const oclMat &img1, const oclMat &img2, const oclMat &weights1, const oclMat &weights2, oclMat &result);
956 //! computes vertical sum, supports only CV_32FC1 images
957 CV_EXPORTS void columnSum(const oclMat &src, oclMat &sum);
959 ///////////////////////////////////////// match_template /////////////////////////////////////////////////////////////
960 struct CV_EXPORTS MatchTemplateBuf
962 Size user_block_size;
963 oclMat imagef, templf;
964 std::vector<oclMat> images;
965 std::vector<oclMat> image_sums;
966 std::vector<oclMat> image_sqsums;
969 //! computes the proximity map for the raster template and the image where the template is searched for
970 // Supports TM_SQDIFF, TM_SQDIFF_NORMED, TM_CCORR, TM_CCORR_NORMED, TM_CCOEFF, TM_CCOEFF_NORMED for type 8UC1 and 8UC4
971 // Supports TM_SQDIFF, TM_CCORR for type 32FC1 and 32FC4
972 CV_EXPORTS void matchTemplate(const oclMat &image, const oclMat &templ, oclMat &result, int method);
974 //! computes the proximity map for the raster template and the image where the template is searched for
975 // Supports TM_SQDIFF, TM_SQDIFF_NORMED, TM_CCORR, TM_CCORR_NORMED, TM_CCOEFF, TM_CCOEFF_NORMED for type 8UC1 and 8UC4
976 // Supports TM_SQDIFF, TM_CCORR for type 32FC1 and 32FC4
977 CV_EXPORTS void matchTemplate(const oclMat &image, const oclMat &templ, oclMat &result, int method, MatchTemplateBuf &buf);
979 ///////////////////////////////////////////// Canny /////////////////////////////////////////////
980 struct CV_EXPORTS CannyBuf;
981 //! compute edges of the input image using Canny operator
982 // Support CV_8UC1 only
983 CV_EXPORTS void Canny(const oclMat &image, oclMat &edges, double low_thresh, double high_thresh, int apperture_size = 3, bool L2gradient = false);
984 CV_EXPORTS void Canny(const oclMat &image, CannyBuf &buf, oclMat &edges, double low_thresh, double high_thresh, int apperture_size = 3, bool L2gradient = false);
985 CV_EXPORTS void Canny(const oclMat &dx, const oclMat &dy, oclMat &edges, double low_thresh, double high_thresh, bool L2gradient = false);
986 CV_EXPORTS void Canny(const oclMat &dx, const oclMat &dy, CannyBuf &buf, oclMat &edges, double low_thresh, double high_thresh, bool L2gradient = false);
988 struct CV_EXPORTS CannyBuf
990 CannyBuf() : counter(NULL) {}
995 explicit CannyBuf(const Size &image_size, int apperture_size = 3) : counter(NULL)
997 create(image_size, apperture_size);
999 CannyBuf(const oclMat &dx_, const oclMat &dy_);
1001 void create(const Size &image_size, int apperture_size = 3);
1004 oclMat dx_buf, dy_buf;
1006 oclMat trackBuf1, trackBuf2;
1008 Ptr<FilterEngine_GPU> filterDX, filterDY;
1011 ///////////////////////////////////////// clAmdFft related /////////////////////////////////////////
1012 //! Performs a forward or inverse discrete Fourier transform (1D or 2D) of floating point matrix.
1013 //! Param dft_size is the size of DFT transform.
1015 //! For complex-to-real transform it is assumed that the source matrix is packed in CLFFT's format.
1016 // support src type of CV32FC1, CV32FC2
1017 // support flags: DFT_INVERSE, DFT_REAL_OUTPUT, DFT_COMPLEX_OUTPUT, DFT_ROWS
1018 // dft_size is the size of original input, which is used for transformation from complex to real.
1019 // dft_size must be powers of 2, 3 and 5
1020 // real to complex dft requires at least v1.8 clAmdFft
1021 // real to complex dft output is not the same with cpu version
1022 // real to complex and complex to real does not support DFT_ROWS
1023 CV_EXPORTS void dft(const oclMat &src, oclMat &dst, Size dft_size = Size(), int flags = 0);
1025 //! implements generalized matrix product algorithm GEMM from BLAS
1026 // The functionality requires clAmdBlas library
1027 // only support type CV_32FC1
1028 // flag GEMM_3_T is not supported
1029 CV_EXPORTS void gemm(const oclMat &src1, const oclMat &src2, double alpha,
1030 const oclMat &src3, double beta, oclMat &dst, int flags = 0);
1032 //////////////// HOG (Histogram-of-Oriented-Gradients) Descriptor and Object Detector //////////////
1033 struct CV_EXPORTS HOGDescriptor
1035 enum { DEFAULT_WIN_SIGMA = -1 };
1036 enum { DEFAULT_NLEVELS = 64 };
1037 enum { DESCR_FORMAT_ROW_BY_ROW, DESCR_FORMAT_COL_BY_COL };
1038 HOGDescriptor(Size win_size = Size(64, 128), Size block_size = Size(16, 16),
1039 Size block_stride = Size(8, 8), Size cell_size = Size(8, 8),
1040 int nbins = 9, double win_sigma = DEFAULT_WIN_SIGMA,
1041 double threshold_L2hys = 0.2, bool gamma_correction = true,
1042 int nlevels = DEFAULT_NLEVELS);
1044 size_t getDescriptorSize() const;
1045 size_t getBlockHistogramSize() const;
1046 void setSVMDetector(const vector<float> &detector);
1047 static vector<float> getDefaultPeopleDetector();
1048 static vector<float> getPeopleDetector48x96();
1049 static vector<float> getPeopleDetector64x128();
1050 void detect(const oclMat &img, vector<Point> &found_locations,
1051 double hit_threshold = 0, Size win_stride = Size(),
1052 Size padding = Size());
1053 void detectMultiScale(const oclMat &img, vector<Rect> &found_locations,
1054 double hit_threshold = 0, Size win_stride = Size(),
1055 Size padding = Size(), double scale0 = 1.05,
1056 int group_threshold = 2);
1057 void getDescriptors(const oclMat &img, Size win_stride,
1058 oclMat &descriptors,
1059 int descr_format = DESCR_FORMAT_COL_BY_COL);
1067 double threshold_L2hys;
1068 bool gamma_correction;
1072 // initialize buffers; only need to do once in case of multiscale detection
1073 void init_buffer(const oclMat &img, Size win_stride);
1074 void computeBlockHistograms(const oclMat &img);
1075 void computeGradient(const oclMat &img, oclMat &grad, oclMat &qangle);
1076 double getWinSigma() const;
1077 bool checkDetectorSize() const;
1079 static int numPartsWithin(int size, int part_size, int stride);
1080 static Size numPartsWithin(Size size, Size part_size, Size stride);
1082 // Coefficients of the separating plane
1085 // Results of the last classification step
1088 // Results of the last histogram evaluation step
1090 // Gradients conputation results
1091 oclMat grad, qangle;
1094 // effect size of input image (might be different from original size after scaling)
1099 ////////////////////////feature2d_ocl/////////////////
1100 /****************************************************************************************\
1102 \****************************************************************************************/
1103 template<typename T>
1104 struct CV_EXPORTS Accumulator
1108 template<> struct Accumulator<unsigned char>
1112 template<> struct Accumulator<unsigned short>
1116 template<> struct Accumulator<char>
1120 template<> struct Accumulator<short>
1126 * Manhattan distance (city block distance) functor
1129 struct CV_EXPORTS L1
1131 enum { normType = NORM_L1 };
1132 typedef T ValueType;
1133 typedef typename Accumulator<T>::Type ResultType;
1135 ResultType operator()( const T *a, const T *b, int size ) const
1137 return normL1<ValueType, ResultType>(a, b, size);
1142 * Euclidean distance functor
1145 struct CV_EXPORTS L2
1147 enum { normType = NORM_L2 };
1148 typedef T ValueType;
1149 typedef typename Accumulator<T>::Type ResultType;
1151 ResultType operator()( const T *a, const T *b, int size ) const
1153 return (ResultType)sqrt((double)normL2Sqr<ValueType, ResultType>(a, b, size));
1158 * Hamming distance functor - counts the bit differences between two strings - useful for the Brief descriptor
1159 * bit count of A exclusive XOR'ed with B
1161 struct CV_EXPORTS Hamming
1163 enum { normType = NORM_HAMMING };
1164 typedef unsigned char ValueType;
1165 typedef int ResultType;
1167 /** this will count the bits in a ^ b
1169 ResultType operator()( const unsigned char *a, const unsigned char *b, int size ) const
1171 return normHamming(a, b, size);
1175 ////////////////////////////////// BruteForceMatcher //////////////////////////////////
1177 class CV_EXPORTS BruteForceMatcher_OCL_base
1180 enum DistType {L1Dist = 0, L2Dist, HammingDist};
1181 explicit BruteForceMatcher_OCL_base(DistType distType = L2Dist);
1182 // Add descriptors to train descriptor collection
1183 void add(const std::vector<oclMat> &descCollection);
1184 // Get train descriptors collection
1185 const std::vector<oclMat> &getTrainDescriptors() const;
1186 // Clear train descriptors collection
1188 // Return true if there are not train descriptors in collection
1191 // Return true if the matcher supports mask in match methods
1192 bool isMaskSupported() const;
1194 // Find one best match for each query descriptor
1195 void matchSingle(const oclMat &query, const oclMat &train,
1196 oclMat &trainIdx, oclMat &distance,
1197 const oclMat &mask = oclMat());
1199 // Download trainIdx and distance and convert it to CPU vector with DMatch
1200 static void matchDownload(const oclMat &trainIdx, const oclMat &distance, std::vector<DMatch> &matches);
1201 // Convert trainIdx and distance to vector with DMatch
1202 static void matchConvert(const Mat &trainIdx, const Mat &distance, std::vector<DMatch> &matches);
1204 // Find one best match for each query descriptor
1205 void match(const oclMat &query, const oclMat &train, std::vector<DMatch> &matches, const oclMat &mask = oclMat());
1207 // Make gpu collection of trains and masks in suitable format for matchCollection function
1208 void makeGpuCollection(oclMat &trainCollection, oclMat &maskCollection, const std::vector<oclMat> &masks = std::vector<oclMat>());
1211 // Find one best match from train collection for each query descriptor
1212 void matchCollection(const oclMat &query, const oclMat &trainCollection,
1213 oclMat &trainIdx, oclMat &imgIdx, oclMat &distance,
1214 const oclMat &masks = oclMat());
1216 // Download trainIdx, imgIdx and distance and convert it to vector with DMatch
1217 static void matchDownload(const oclMat &trainIdx, const oclMat &imgIdx, const oclMat &distance, std::vector<DMatch> &matches);
1218 // Convert trainIdx, imgIdx and distance to vector with DMatch
1219 static void matchConvert(const Mat &trainIdx, const Mat &imgIdx, const Mat &distance, std::vector<DMatch> &matches);
1221 // Find one best match from train collection for each query descriptor.
1222 void match(const oclMat &query, std::vector<DMatch> &matches, const std::vector<oclMat> &masks = std::vector<oclMat>());
1224 // Find k best matches for each query descriptor (in increasing order of distances)
1225 void knnMatchSingle(const oclMat &query, const oclMat &train,
1226 oclMat &trainIdx, oclMat &distance, oclMat &allDist, int k,
1227 const oclMat &mask = oclMat());
1229 // Download trainIdx and distance and convert it to vector with DMatch
1230 // compactResult is used when mask is not empty. If compactResult is false matches
1231 // vector will have the same size as queryDescriptors rows. If compactResult is true
1232 // matches vector will not contain matches for fully masked out query descriptors.
1233 static void knnMatchDownload(const oclMat &trainIdx, const oclMat &distance,
1234 std::vector< std::vector<DMatch> > &matches, bool compactResult = false);
1236 // Convert trainIdx and distance to vector with DMatch
1237 static void knnMatchConvert(const Mat &trainIdx, const Mat &distance,
1238 std::vector< std::vector<DMatch> > &matches, bool compactResult = false);
1240 // Find k best matches for each query descriptor (in increasing order of distances).
1241 // compactResult is used when mask is not empty. If compactResult is false matches
1242 // vector will have the same size as queryDescriptors rows. If compactResult is true
1243 // matches vector will not contain matches for fully masked out query descriptors.
1244 void knnMatch(const oclMat &query, const oclMat &train,
1245 std::vector< std::vector<DMatch> > &matches, int k, const oclMat &mask = oclMat(),
1246 bool compactResult = false);
1248 // Find k best matches from train collection for each query descriptor (in increasing order of distances)
1249 void knnMatch2Collection(const oclMat &query, const oclMat &trainCollection,
1250 oclMat &trainIdx, oclMat &imgIdx, oclMat &distance,
1251 const oclMat &maskCollection = oclMat());
1253 // Download trainIdx and distance and convert it to vector with DMatch
1254 // compactResult is used when mask is not empty. If compactResult is false matches
1255 // vector will have the same size as queryDescriptors rows. If compactResult is true
1256 // matches vector will not contain matches for fully masked out query descriptors.
1257 static void knnMatch2Download(const oclMat &trainIdx, const oclMat &imgIdx, const oclMat &distance,
1258 std::vector< std::vector<DMatch> > &matches, bool compactResult = false);
1260 // Convert trainIdx and distance to vector with DMatch
1261 static void knnMatch2Convert(const Mat &trainIdx, const Mat &imgIdx, const Mat &distance,
1262 std::vector< std::vector<DMatch> > &matches, bool compactResult = false);
1264 // Find k best matches for each query descriptor (in increasing order of distances).
1265 // compactResult is used when mask is not empty. If compactResult is false matches
1266 // vector will have the same size as queryDescriptors rows. If compactResult is true
1267 // matches vector will not contain matches for fully masked out query descriptors.
1268 void knnMatch(const oclMat &query, std::vector< std::vector<DMatch> > &matches, int k,
1269 const std::vector<oclMat> &masks = std::vector<oclMat>(), bool compactResult = false);
1271 // Find best matches for each query descriptor which have distance less than maxDistance.
1272 // nMatches.at<int>(0, queryIdx) will contain matches count for queryIdx.
1273 // carefully nMatches can be greater than trainIdx.cols - it means that matcher didn't find all matches,
1274 // because it didn't have enough memory.
1275 // If trainIdx is empty, then trainIdx and distance will be created with size nQuery x max((nTrain / 100), 10),
1276 // otherwize user can pass own allocated trainIdx and distance with size nQuery x nMaxMatches
1277 // Matches doesn't sorted.
1278 void radiusMatchSingle(const oclMat &query, const oclMat &train,
1279 oclMat &trainIdx, oclMat &distance, oclMat &nMatches, float maxDistance,
1280 const oclMat &mask = oclMat());
1282 // Download trainIdx, nMatches and distance and convert it to vector with DMatch.
1283 // matches will be sorted in increasing order of distances.
1284 // compactResult is used when mask is not empty. If compactResult is false matches
1285 // vector will have the same size as queryDescriptors rows. If compactResult is true
1286 // matches vector will not contain matches for fully masked out query descriptors.
1287 static void radiusMatchDownload(const oclMat &trainIdx, const oclMat &distance, const oclMat &nMatches,
1288 std::vector< std::vector<DMatch> > &matches, bool compactResult = false);
1289 // Convert trainIdx, nMatches and distance to vector with DMatch.
1290 static void radiusMatchConvert(const Mat &trainIdx, const Mat &distance, const Mat &nMatches,
1291 std::vector< std::vector<DMatch> > &matches, bool compactResult = false);
1292 // Find best matches for each query descriptor which have distance less than maxDistance
1293 // in increasing order of distances).
1294 void radiusMatch(const oclMat &query, const oclMat &train,
1295 std::vector< std::vector<DMatch> > &matches, float maxDistance,
1296 const oclMat &mask = oclMat(), bool compactResult = false);
1297 // Find best matches for each query descriptor which have distance less than maxDistance.
1298 // If trainIdx is empty, then trainIdx and distance will be created with size nQuery x max((nQuery / 100), 10),
1299 // otherwize user can pass own allocated trainIdx and distance with size nQuery x nMaxMatches
1300 // Matches doesn't sorted.
1301 void radiusMatchCollection(const oclMat &query, oclMat &trainIdx, oclMat &imgIdx, oclMat &distance, oclMat &nMatches, float maxDistance,
1302 const std::vector<oclMat> &masks = std::vector<oclMat>());
1303 // Download trainIdx, imgIdx, nMatches and distance and convert it to vector with DMatch.
1304 // matches will be sorted in increasing order of distances.
1305 // compactResult is used when mask is not empty. If compactResult is false matches
1306 // vector will have the same size as queryDescriptors rows. If compactResult is true
1307 // matches vector will not contain matches for fully masked out query descriptors.
1308 static void radiusMatchDownload(const oclMat &trainIdx, const oclMat &imgIdx, const oclMat &distance, const oclMat &nMatches,
1309 std::vector< std::vector<DMatch> > &matches, bool compactResult = false);
1310 // Convert trainIdx, nMatches and distance to vector with DMatch.
1311 static void radiusMatchConvert(const Mat &trainIdx, const Mat &imgIdx, const Mat &distance, const Mat &nMatches,
1312 std::vector< std::vector<DMatch> > &matches, bool compactResult = false);
1313 // Find best matches from train collection for each query descriptor which have distance less than
1314 // maxDistance (in increasing order of distances).
1315 void radiusMatch(const oclMat &query, std::vector< std::vector<DMatch> > &matches, float maxDistance,
1316 const std::vector<oclMat> &masks = std::vector<oclMat>(), bool compactResult = false);
1319 std::vector<oclMat> trainDescCollection;
1322 template <class Distance>
1323 class CV_EXPORTS BruteForceMatcher_OCL;
1325 template <typename T>
1326 class CV_EXPORTS BruteForceMatcher_OCL< L1<T> > : public BruteForceMatcher_OCL_base
1329 explicit BruteForceMatcher_OCL() : BruteForceMatcher_OCL_base(L1Dist) {}
1330 explicit BruteForceMatcher_OCL(L1<T> /*d*/) : BruteForceMatcher_OCL_base(L1Dist) {}
1333 template <typename T>
1334 class CV_EXPORTS BruteForceMatcher_OCL< L2<T> > : public BruteForceMatcher_OCL_base
1337 explicit BruteForceMatcher_OCL() : BruteForceMatcher_OCL_base(L2Dist) {}
1338 explicit BruteForceMatcher_OCL(L2<T> /*d*/) : BruteForceMatcher_OCL_base(L2Dist) {}
1341 template <> class CV_EXPORTS BruteForceMatcher_OCL< Hamming > : public BruteForceMatcher_OCL_base
1344 explicit BruteForceMatcher_OCL() : BruteForceMatcher_OCL_base(HammingDist) {}
1345 explicit BruteForceMatcher_OCL(Hamming /*d*/) : BruteForceMatcher_OCL_base(HammingDist) {}
1348 class CV_EXPORTS BFMatcher_OCL : public BruteForceMatcher_OCL_base
1351 explicit BFMatcher_OCL(int norm = NORM_L2) : BruteForceMatcher_OCL_base(norm == NORM_L1 ? L1Dist : norm == NORM_L2 ? L2Dist : HammingDist) {}
1354 class CV_EXPORTS GoodFeaturesToTrackDetector_OCL
1357 explicit GoodFeaturesToTrackDetector_OCL(int maxCorners = 1000, double qualityLevel = 0.01, double minDistance = 0.0,
1358 int blockSize = 3, bool useHarrisDetector = false, double harrisK = 0.04);
1360 //! return 1 rows matrix with CV_32FC2 type
1361 void operator ()(const oclMat& image, oclMat& corners, const oclMat& mask = oclMat());
1362 //! download points of type Point2f to a vector. the vector's content will be erased
1363 void downloadPoints(const oclMat &points, vector<Point2f> &points_v);
1366 double qualityLevel;
1370 bool useHarrisDetector;
1372 void releaseMemory()
1377 minMaxbuf_.release();
1378 tmpCorners_.release();
1388 inline GoodFeaturesToTrackDetector_OCL::GoodFeaturesToTrackDetector_OCL(int maxCorners_, double qualityLevel_, double minDistance_,
1389 int blockSize_, bool useHarrisDetector_, double harrisK_)
1391 maxCorners = maxCorners_;
1392 qualityLevel = qualityLevel_;
1393 minDistance = minDistance_;
1394 blockSize = blockSize_;
1395 useHarrisDetector = useHarrisDetector_;
1399 /////////////////////////////// PyrLKOpticalFlow /////////////////////////////////////
1400 class CV_EXPORTS PyrLKOpticalFlow
1405 winSize = Size(21, 21);
1409 useInitialFlow = false;
1410 minEigThreshold = 1e-4f;
1411 getMinEigenVals = false;
1412 isDeviceArch11_ = false;
1415 void sparse(const oclMat &prevImg, const oclMat &nextImg, const oclMat &prevPts, oclMat &nextPts,
1416 oclMat &status, oclMat *err = 0);
1417 void dense(const oclMat &prevImg, const oclMat &nextImg, oclMat &u, oclMat &v, oclMat *err = 0);
1422 bool useInitialFlow;
1423 float minEigThreshold;
1424 bool getMinEigenVals;
1425 void releaseMemory()
1427 dx_calcBuf_.release();
1428 dy_calcBuf_.release();
1437 void calcSharrDeriv(const oclMat &src, oclMat &dx, oclMat &dy);
1438 void buildImagePyramid(const oclMat &img0, vector<oclMat> &pyr, bool withBorder);
1443 vector<oclMat> prevPyr_;
1444 vector<oclMat> nextPyr_;
1450 bool isDeviceArch11_;
1453 class CV_EXPORTS FarnebackOpticalFlow
1456 FarnebackOpticalFlow();
1467 void operator ()(const oclMat &frame0, const oclMat &frame1, oclMat &flowx, oclMat &flowy);
1469 void releaseMemory();
1472 void prepareGaussian(
1473 int n, double sigma, float *g, float *xg, float *xxg,
1474 double &ig11, double &ig03, double &ig33, double &ig55);
1476 void setPolynomialExpansionConsts(int n, double sigma);
1478 void updateFlow_boxFilter(
1479 const oclMat& R0, const oclMat& R1, oclMat& flowx, oclMat &flowy,
1480 oclMat& M, oclMat &bufM, int blockSize, bool updateMatrices);
1482 void updateFlow_gaussianBlur(
1483 const oclMat& R0, const oclMat& R1, oclMat& flowx, oclMat& flowy,
1484 oclMat& M, oclMat &bufM, int blockSize, bool updateMatrices);
1487 oclMat pyrLevel_[2], M_, bufM_, R_[2], blurredFrame_[2];
1488 std::vector<oclMat> pyramid0_, pyramid1_;
1491 //////////////// build warping maps ////////////////////
1492 //! builds plane warping maps
1493 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);
1494 //! builds cylindrical warping maps
1495 CV_EXPORTS void buildWarpCylindricalMaps(Size src_size, Rect dst_roi, const Mat &K, const Mat &R, float scale, oclMat &map_x, oclMat &map_y);
1496 //! builds spherical warping maps
1497 CV_EXPORTS void buildWarpSphericalMaps(Size src_size, Rect dst_roi, const Mat &K, const Mat &R, float scale, oclMat &map_x, oclMat &map_y);
1498 //! builds Affine warping maps
1499 CV_EXPORTS void buildWarpAffineMaps(const Mat &M, bool inverse, Size dsize, oclMat &xmap, oclMat &ymap);
1501 //! builds Perspective warping maps
1502 CV_EXPORTS void buildWarpPerspectiveMaps(const Mat &M, bool inverse, Size dsize, oclMat &xmap, oclMat &ymap);
1504 ///////////////////////////////////// interpolate frames //////////////////////////////////////////////
1505 //! Interpolate frames (images) using provided optical flow (displacement field).
1506 //! frame0 - frame 0 (32-bit floating point images, single channel)
1507 //! frame1 - frame 1 (the same type and size)
1508 //! fu - forward horizontal displacement
1509 //! fv - forward vertical displacement
1510 //! bu - backward horizontal displacement
1511 //! bv - backward vertical displacement
1512 //! pos - new frame position
1513 //! newFrame - new frame
1514 //! buf - temporary buffer, will have width x 6*height size, CV_32FC1 type and contain 6 oclMat;
1515 //! occlusion masks 0, occlusion masks 1,
1516 //! interpolated forward flow 0, interpolated forward flow 1,
1517 //! interpolated backward flow 0, interpolated backward flow 1
1519 CV_EXPORTS void interpolateFrames(const oclMat &frame0, const oclMat &frame1,
1520 const oclMat &fu, const oclMat &fv,
1521 const oclMat &bu, const oclMat &bv,
1522 float pos, oclMat &newFrame, oclMat &buf);
1524 //! computes moments of the rasterized shape or a vector of points
1525 CV_EXPORTS Moments ocl_moments(InputArray _array, bool binaryImage);
1527 class CV_EXPORTS StereoBM_OCL
1530 enum { BASIC_PRESET = 0, PREFILTER_XSOBEL = 1 };
1532 enum { DEFAULT_NDISP = 64, DEFAULT_WINSZ = 19 };
1534 //! the default constructor
1536 //! the full constructor taking the camera-specific preset, number of disparities and the SAD window size. ndisparities must be multiple of 8.
1537 StereoBM_OCL(int preset, int ndisparities = DEFAULT_NDISP, int winSize = DEFAULT_WINSZ);
1539 //! the stereo correspondence operator. Finds the disparity for the specified rectified stereo pair
1540 //! Output disparity has CV_8U type.
1541 void operator() ( const oclMat &left, const oclMat &right, oclMat &disparity);
1543 //! Some heuristics that tries to estmate
1544 // if current GPU will be faster then CPU in this algorithm.
1545 // It queries current active device.
1546 static bool checkIfGpuCallReasonable();
1552 // If avergeTexThreshold == 0 => post procesing is disabled
1553 // If avergeTexThreshold != 0 then disparity is set 0 in each point (x,y) where for left image
1554 // SumOfHorizontalGradiensInWindow(x, y, winSize) < (winSize * winSize) * avergeTexThreshold
1555 // i.e. input left image is low textured.
1556 float avergeTexThreshold;
1558 oclMat minSSD, leBuf, riBuf;
1561 class CV_EXPORTS StereoBeliefPropagation
1564 enum { DEFAULT_NDISP = 64 };
1565 enum { DEFAULT_ITERS = 5 };
1566 enum { DEFAULT_LEVELS = 5 };
1567 static void estimateRecommendedParams(int width, int height, int &ndisp, int &iters, int &levels);
1568 explicit StereoBeliefPropagation(int ndisp = DEFAULT_NDISP,
1569 int iters = DEFAULT_ITERS,
1570 int levels = DEFAULT_LEVELS,
1571 int msg_type = CV_16S);
1572 StereoBeliefPropagation(int ndisp, int iters, int levels,
1573 float max_data_term, float data_weight,
1574 float max_disc_term, float disc_single_jump,
1575 int msg_type = CV_32F);
1576 void operator()(const oclMat &left, const oclMat &right, oclMat &disparity);
1577 void operator()(const oclMat &data, oclMat &disparity);
1581 float max_data_term;
1583 float max_disc_term;
1584 float disc_single_jump;
1587 oclMat u, d, l, r, u2, d2, l2, r2;
1588 std::vector<oclMat> datas;
1592 class CV_EXPORTS StereoConstantSpaceBP
1595 enum { DEFAULT_NDISP = 128 };
1596 enum { DEFAULT_ITERS = 8 };
1597 enum { DEFAULT_LEVELS = 4 };
1598 enum { DEFAULT_NR_PLANE = 4 };
1599 static void estimateRecommendedParams(int width, int height, int &ndisp, int &iters, int &levels, int &nr_plane);
1600 explicit StereoConstantSpaceBP(
1601 int ndisp = DEFAULT_NDISP,
1602 int iters = DEFAULT_ITERS,
1603 int levels = DEFAULT_LEVELS,
1604 int nr_plane = DEFAULT_NR_PLANE,
1605 int msg_type = CV_32F);
1606 StereoConstantSpaceBP(int ndisp, int iters, int levels, int nr_plane,
1607 float max_data_term, float data_weight, float max_disc_term, float disc_single_jump,
1608 int min_disp_th = 0,
1609 int msg_type = CV_32F);
1610 void operator()(const oclMat &left, const oclMat &right, oclMat &disparity);
1615 float max_data_term;
1617 float max_disc_term;
1618 float disc_single_jump;
1621 bool use_local_init_data_cost;
1623 oclMat u[2], d[2], l[2], r[2];
1624 oclMat disp_selected_pyr[2];
1626 oclMat data_cost_selected;
1631 // Implementation of the Zach, Pock and Bischof Dual TV-L1 Optical Flow method
1634 // [1] C. Zach, T. Pock and H. Bischof, "A Duality Based Approach for Realtime TV-L1 Optical Flow".
1635 // [2] Javier Sanchez, Enric Meinhardt-Llopis and Gabriele Facciolo. "TV-L1 Optical Flow Estimation".
1636 class CV_EXPORTS OpticalFlowDual_TVL1_OCL
1639 OpticalFlowDual_TVL1_OCL();
1641 void operator ()(const oclMat& I0, const oclMat& I1, oclMat& flowx, oclMat& flowy);
1643 void collectGarbage();
1646 * Time step of the numerical scheme.
1651 * Weight parameter for the data term, attachment parameter.
1652 * This is the most relevant parameter, which determines the smoothness of the output.
1653 * The smaller this parameter is, the smoother the solutions we obtain.
1654 * It depends on the range of motions of the images, so its value should be adapted to each image sequence.
1659 * Weight parameter for (u - v)^2, tightness parameter.
1660 * It serves as a link between the attachment and the regularization terms.
1661 * In theory, it should have a small value in order to maintain both parts in correspondence.
1662 * The method is stable for a large range of values of this parameter.
1667 * Number of scales used to create the pyramid of images.
1672 * Number of warpings per scale.
1673 * Represents the number of times that I1(x+u0) and grad( I1(x+u0) ) are computed per scale.
1674 * This is a parameter that assures the stability of the method.
1675 * It also affects the running time, so it is a compromise between speed and accuracy.
1680 * Stopping criterion threshold used in the numerical scheme, which is a trade-off between precision and running time.
1681 * A small value will yield more accurate solutions at the expense of a slower convergence.
1686 * Stopping criterion iterations number used in the numerical scheme.
1690 bool useInitialFlow;
1693 void procOneScale(const oclMat& I0, const oclMat& I1, oclMat& u1, oclMat& u2);
1695 std::vector<oclMat> I0s;
1696 std::vector<oclMat> I1s;
1697 std::vector<oclMat> u1s;
1698 std::vector<oclMat> u2s;
1718 // current supported sorting methods
1721 SORT_BITONIC, // only support power-of-2 buffer size
1722 SORT_SELECTION, // cannot sort duplicate keys
1724 SORT_RADIX // only support signed int/float keys(CV_32S/CV_32F)
1726 //! Returns the sorted result of all the elements in input based on equivalent keys.
1728 // The element unit in the values to be sorted is determined from the data type,
1729 // i.e., a CV_32FC2 input {a1a2, b1b2} will be considered as two elements, regardless its
1730 // matrix dimension.
1731 // both keys and values will be sorted inplace
1732 // Key needs to be single channel oclMat.
1736 // keys = {2, 3, 1} (CV_8UC1)
1737 // values = {10,5, 4,3, 6,2} (CV_8UC2)
1738 // sortByKey(keys, values, SORT_SELECTION, false);
1740 // keys = {1, 2, 3} (CV_8UC1)
1741 // values = {6,2, 10,5, 4,3} (CV_8UC2)
1742 CV_EXPORTS void sortByKey(oclMat& keys, oclMat& values, int method, bool isGreaterThan = false);
1743 /*!Base class for MOG and MOG2!*/
1744 class CV_EXPORTS BackgroundSubtractor
1747 //! the virtual destructor
1748 virtual ~BackgroundSubtractor();
1749 //! the update operator that takes the next video frame and returns the current foreground mask as 8-bit binary image.
1750 virtual void operator()(const oclMat& image, oclMat& fgmask, float learningRate);
1752 //! computes a background image
1753 virtual void getBackgroundImage(oclMat& backgroundImage) const = 0;
1756 Gaussian Mixture-based Backbround/Foreground Segmentation Algorithm
1758 The class implements the following algorithm:
1759 "An improved adaptive background mixture model for real-time tracking with shadow detection"
1760 P. KadewTraKuPong and R. Bowden,
1761 Proc. 2nd European Workshp on Advanced Video-Based Surveillance Systems, 2001."
1762 http://personal.ee.surrey.ac.uk/Personal/R.Bowden/publications/avbs01/avbs01.pdf
1764 class CV_EXPORTS MOG: public cv::ocl::BackgroundSubtractor
1767 //! the default constructor
1768 MOG(int nmixtures = -1);
1770 //! re-initiaization method
1771 void initialize(Size frameSize, int frameType);
1773 //! the update operator
1774 void operator()(const oclMat& frame, oclMat& fgmask, float learningRate = 0.f);
1776 //! computes a background image which are the mean of all background gaussians
1777 void getBackgroundImage(oclMat& backgroundImage) const;
1779 //! releases all inner buffers
1784 float backgroundRatio;
1801 The class implements the following algorithm:
1802 "Improved adaptive Gausian mixture model for background subtraction"
1804 International Conference Pattern Recognition, UK, August, 2004.
1805 http://www.zoranz.net/Publications/zivkovic2004ICPR.pdf
1807 class CV_EXPORTS MOG2: public cv::ocl::BackgroundSubtractor
1810 //! the default constructor
1811 MOG2(int nmixtures = -1);
1813 //! re-initiaization method
1814 void initialize(Size frameSize, int frameType);
1816 //! the update operator
1817 void operator()(const oclMat& frame, oclMat& fgmask, float learningRate = -1.0f);
1819 //! computes a background image which are the mean of all background gaussians
1820 void getBackgroundImage(oclMat& backgroundImage) const;
1822 //! releases all inner buffers
1826 // you should call initialize after parameters changes
1830 //! here it is the maximum allowed number of mixture components.
1831 //! Actual number is determined dynamically per pixel
1833 // threshold on the squared Mahalanobis distance to decide if it is well described
1834 // by the background model or not. Related to Cthr from the paper.
1835 // This does not influence the update of the background. A typical value could be 4 sigma
1836 // and that is varThreshold=4*4=16; Corresponds to Tb in the paper.
1838 /////////////////////////
1839 // less important parameters - things you might change but be carefull
1840 ////////////////////////
1842 float backgroundRatio;
1843 // corresponds to fTB=1-cf from the paper
1844 // TB - threshold when the component becomes significant enough to be included into
1845 // the background model. It is the TB=1-cf from the paper. So I use cf=0.1 => TB=0.
1846 // For alpha=0.001 it means that the mode should exist for approximately 105 frames before
1847 // it is considered foreground
1848 // float noiseSigma;
1849 float varThresholdGen;
1851 //correspondts to Tg - threshold on the squared Mahalan. dist. to decide
1852 //when a sample is close to the existing components. If it is not close
1853 //to any a new component will be generated. I use 3 sigma => Tg=3*3=9.
1854 //Smaller Tg leads to more generated components and higher Tg might make
1855 //lead to small number of components but they can grow too large
1860 //initial variance for the newly generated components.
1861 //It will will influence the speed of adaptation. A good guess should be made.
1862 //A simple way is to estimate the typical standard deviation from the images.
1863 //I used here 10 as a reasonable value
1864 // min and max can be used to further control the variance
1865 float fCT; //CT - complexity reduction prior
1866 //this is related to the number of samples needed to accept that a component
1867 //actually exists. We use CT=0.05 of all the samples. By setting CT=0 you get
1868 //the standard Stauffer&Grimson algorithm (maybe not exact but very similar)
1870 //shadow detection parameters
1871 bool bShadowDetection; //default 1 - do shadow detection
1872 unsigned char nShadowDetection; //do shadow detection - insert this value as the detection result - 127 default value
1874 // Tau - shadow threshold. The shadow is detected if the pixel is darker
1875 //version of the background. Tau is a threshold on how much darker the shadow can be.
1876 //Tau= 0.5 means that if pixel is more than 2 times darker then it is not shadow
1877 //See: Prati,Mikic,Trivedi,Cucchiarra,"Detecting Moving Shadows...",IEEE PAMI,2003.
1890 oclMat bgmodelUsedModes_; //keep track of number of modes per pixel
1893 /*!***************Kalman Filter*************!*/
1894 class CV_EXPORTS KalmanFilter
1898 //! the full constructor taking the dimensionality of the state, of the measurement and of the control vector
1899 KalmanFilter(int dynamParams, int measureParams, int controlParams=0, int type=CV_32F);
1900 //! re-initializes Kalman filter. The previous content is destroyed.
1901 void init(int dynamParams, int measureParams, int controlParams=0, int type=CV_32F);
1903 const oclMat& predict(const oclMat& control=oclMat());
1904 const oclMat& correct(const oclMat& measurement);
1906 oclMat statePre; //!< predicted state (x'(k)): x(k)=A*x(k-1)+B*u(k)
1907 oclMat statePost; //!< corrected state (x(k)): x(k)=x'(k)+K(k)*(z(k)-H*x'(k))
1908 oclMat transitionMatrix; //!< state transition matrix (A)
1909 oclMat controlMatrix; //!< control matrix (B) (not used if there is no control)
1910 oclMat measurementMatrix; //!< measurement matrix (H)
1911 oclMat processNoiseCov; //!< process noise covariance matrix (Q)
1912 oclMat measurementNoiseCov;//!< measurement noise covariance matrix (R)
1913 oclMat errorCovPre; //!< priori error estimate covariance matrix (P'(k)): P'(k)=A*P(k-1)*At + Q)*/
1914 oclMat gain; //!< Kalman gain matrix (K(k)): K(k)=P'(k)*Ht*inv(H*P'(k)*Ht+R)
1915 oclMat errorCovPost; //!< posteriori error estimate covariance matrix (P(k)): P(k)=(I-K(k)*H)*P'(k)
1924 /*!***************K Nearest Neighbour*************!*/
1925 class CV_EXPORTS KNearestNeighbour: public CvKNearest
1928 KNearestNeighbour();
1929 ~KNearestNeighbour();
1931 bool train(const Mat& trainData, Mat& labels, Mat& sampleIdx = Mat().setTo(Scalar::all(0)),
1932 bool isRegression = false, int max_k = 32, bool updateBase = false);
1936 void find_nearest(const oclMat& samples, int k, oclMat& lables);
1942 /*!*************** SVM *************!*/
1943 class CV_EXPORTS CvSVM_OCL : public CvSVM
1948 CvSVM_OCL(const cv::Mat& trainData, const cv::Mat& responses,
1949 const cv::Mat& varIdx=cv::Mat(), const cv::Mat& sampleIdx=cv::Mat(),
1950 CvSVMParams params=CvSVMParams());
1951 CV_WRAP float predict( const int row_index, Mat& src, bool returnDFVal=false ) const;
1952 CV_WRAP void predict( cv::InputArray samples, cv::OutputArray results ) const;
1953 CV_WRAP float predict( const cv::Mat& sample, bool returnDFVal=false ) const;
1954 float predict( const CvMat* samples, CV_OUT CvMat* results ) const;
1957 float predict( const int row_index, int row_len, Mat& src, bool returnDFVal=false ) const;
1958 void create_kernel();
1959 void create_solver();
1962 /*!*************** END *************!*/
1965 #if defined _MSC_VER && _MSC_VER >= 1200
1966 # pragma warning( push)
1967 # pragma warning( disable: 4267)
1969 #include "opencv2/ocl/matrix_operations.hpp"
1970 #if defined _MSC_VER && _MSC_VER >= 1200
1971 # pragma warning( pop)
1974 #endif /* __OPENCV_OCL_HPP__ */