diff --git a/hal/ipp/CMakeLists.txt b/hal/ipp/CMakeLists.txt index 58ae761d62..f4cae58364 100644 --- a/hal/ipp/CMakeLists.txt +++ b/hal/ipp/CMakeLists.txt @@ -22,6 +22,7 @@ add_library(ipphal STATIC "${CMAKE_CURRENT_SOURCE_DIR}/src/sum_ipp.cpp" "${CMAKE_CURRENT_SOURCE_DIR}/src/color_ipp.cpp" "${CMAKE_CURRENT_SOURCE_DIR}/src/filter_ipp.cpp" + "${CMAKE_CURRENT_SOURCE_DIR}/src/matchtemplate_ipp.cpp" ) #TODO: HAVE_IPP_ICV and HAVE_IPP_IW added as private macro till OpenCV itself is diff --git a/hal/ipp/include/ipp_hal_imgproc.hpp b/hal/ipp/include/ipp_hal_imgproc.hpp index 70c643b830..856ea5d222 100644 --- a/hal/ipp/include/ipp_hal_imgproc.hpp +++ b/hal/ipp/include/ipp_hal_imgproc.hpp @@ -138,6 +138,12 @@ int ipp_hal_cvtRGBAtoMultipliedRGBA(const uchar * src_data, size_t src_step, uch #undef cv_hal_cvtRGBAtoMultipliedRGBA #define cv_hal_cvtRGBAtoMultipliedRGBA ipp_hal_cvtRGBAtoMultipliedRGBA +int ipp_hal_matchTemplate(const uchar* src_data, size_t src_step, int src_width, int src_height, + const uchar* templ_data, size_t templ_step, int templ_width, int templ_height, + float* result_data, size_t result_step, int depth, int cn, int method); +#undef cv_hal_matchTemplate +#define cv_hal_matchTemplate ipp_hal_matchTemplate + #endif // IPP_VERSION_X100 >= 700 #endif //__IPP_HAL_IMGPROC_HPP__ diff --git a/hal/ipp/src/matchtemplate_ipp.cpp b/hal/ipp/src/matchtemplate_ipp.cpp new file mode 100644 index 0000000000..7ff0807aa6 --- /dev/null +++ b/hal/ipp/src/matchtemplate_ipp.cpp @@ -0,0 +1,144 @@ +// This file is part of OpenCV project. +// It is subject to the license terms in the LICENSE file found in the top-level directory +// of this distribution and at http://opencv.org/license.html +// Copyright (C) 2026, BigVision LLC, all rights reserved. +// Third party copyrights are property of their respective owners. + +#include "ipp_hal_imgproc.hpp" + +#include +#include "precomp_ipp.hpp" + +#if IPP_VERSION_X100 >= 700 + +using namespace cv; + +#if IPP_VERSION_X100 >= 201700 +#define IPP_HAL_MALLOC(SIZE) ippMalloc_L(SIZE) +#else +#define IPP_HAL_MALLOC(SIZE) ippMalloc((int)(SIZE)) +#endif + +typedef IppStatus (CV_STDCALL * ippimatchTemplate)(const void*, int, IppiSize, const void*, int, IppiSize, Ipp32f* , int , IppEnum , Ipp8u*); + +static bool ipp_crossCorr(const Mat& src, const Mat& tpl, Mat& dst, bool normed) +{ + IppStatus status; + + IppiSize srcRoiSize = {src.cols,src.rows}; + IppiSize tplRoiSize = {tpl.cols,tpl.rows}; + + int bufSize=0; + + int depth = src.depth(); + + ippimatchTemplate ippiCrossCorrNorm = + depth==CV_8U ? (ippimatchTemplate)ippiCrossCorrNorm_8u32f_C1R: + depth==CV_32F? (ippimatchTemplate)ippiCrossCorrNorm_32f_C1R: 0; + + if (ippiCrossCorrNorm==0) + return false; + + IppEnum funCfg = (IppEnum)(+ippAlgAuto | ippiROIValid); + if(normed) + funCfg |= ippiNorm; + else + funCfg |= ippiNormNone; + + status = ippiCrossCorrNormGetBufferSize(srcRoiSize, tplRoiSize, funCfg, &bufSize); + if ( status < 0 ) + return false; + + Ipp8u* buffer = bufSize > 0 ? (Ipp8u*)IPP_HAL_MALLOC(bufSize) : 0; + if (bufSize > 0 && buffer == 0) + return false; + + status = CV_INSTRUMENT_FUN_IPP(ippiCrossCorrNorm, src.ptr(), (int)src.step, srcRoiSize, tpl.ptr(), (int)tpl.step, tplRoiSize, dst.ptr(), (int)dst.step, funCfg, buffer); + + if (buffer) + ippFree(buffer); + return status >= 0; +} + +static bool ipp_sqrDistance(const Mat& src, const Mat& tpl, Mat& dst) +{ + IppStatus status; + + IppiSize srcRoiSize = {src.cols,src.rows}; + IppiSize tplRoiSize = {tpl.cols,tpl.rows}; + + int bufSize=0; + + int depth = src.depth(); + + ippimatchTemplate ippiSqrDistanceNorm = + depth==CV_8U ? (ippimatchTemplate)ippiSqrDistanceNorm_8u32f_C1R: + depth==CV_32F? (ippimatchTemplate)ippiSqrDistanceNorm_32f_C1R: 0; + + if (ippiSqrDistanceNorm==0) + return false; + + IppEnum funCfg = (IppEnum)(+ippAlgAuto | ippiROIValid | ippiNormNone); + status = ippiSqrDistanceNormGetBufferSize(srcRoiSize, tplRoiSize, funCfg, &bufSize); + if ( status < 0 ) + return false; + + Ipp8u* buffer = bufSize > 0 ? (Ipp8u*)IPP_HAL_MALLOC(bufSize) : 0; + if (bufSize > 0 && buffer == 0) + return false; + + status = CV_INSTRUMENT_FUN_IPP(ippiSqrDistanceNorm, src.ptr(), (int)src.step, srcRoiSize, tpl.ptr(), (int)tpl.step, tplRoiSize, dst.ptr(), (int)dst.step, funCfg, buffer); + + if (buffer) + ippFree(buffer); + dst = cv::max(dst, 0); // handle edge case from rounding in variance computation which can result in negative values + return status >= 0; +} + +int ipp_hal_matchTemplate(const uchar* src_data, size_t src_step, int src_width, int src_height, + const uchar* templ_data, size_t templ_step, int templ_width, int templ_height, + float* result_data, size_t result_step, int depth, int cn, int method) +{ + CV_HAL_CHECK_USE_IPP(); + + if(cn != 1) + return CV_HAL_ERROR_NOT_IMPLEMENTED; + if(depth != CV_8U && depth != CV_32F) + return CV_HAL_ERROR_NOT_IMPLEMENTED; + + Mat img(src_height, src_width, CV_MAKETYPE(depth, cn), (void*)src_data, src_step); + Mat templ(templ_height, templ_width, CV_MAKETYPE(depth, cn), (void*)templ_data, templ_step); + Mat result(src_height - templ_height + 1, src_width - templ_width + 1, CV_32FC1, (void*)result_data, result_step); + + // These functions are not efficient if template size is comparable with image size + if(templ.size().area()*4 > img.size().area()) + return CV_HAL_ERROR_NOT_IMPLEMENTED; + + // CV_8U SQDIFF/SQDIFF_NORMED suffer from float32 catastrophic cancellation + // in IPP's internal accumulators; fall through to the double-precision path instead. + if(depth == CV_8U && (method == cv::TM_SQDIFF || method == cv::TM_SQDIFF_NORMED)) + return CV_HAL_ERROR_NOT_IMPLEMENTED; + + if(method == cv::TM_SQDIFF) + { + if(ipp_sqrDistance(img, templ, result)) + return CV_HAL_ERROR_OK; + } + else if(method == cv::TM_CCORR_NORMED) + { + if(ipp_crossCorr(img, templ, result, true)) + return CV_HAL_ERROR_OK; + } + else if(method == cv::TM_SQDIFF_NORMED || method == cv::TM_CCORR || + method == cv::TM_CCOEFF || method == cv::TM_CCOEFF_NORMED) + { + // Raw cross-correlation; caller finishes SQDIFF_NORMED/CCOEFF/CCOEFF_NORMED via + // common_matchTemplate (TM_CCORR is already final). + if(ipp_crossCorr(img, templ, result, false)) + return CV_HAL_ERROR_OK; + } + + return CV_HAL_ERROR_NOT_IMPLEMENTED; +} + +#endif // IPP_VERSION_X100 >= 700 diff --git a/modules/imgproc/src/hal_replacement.hpp b/modules/imgproc/src/hal_replacement.hpp index 4d3fe4477d..f642a3709b 100644 --- a/modules/imgproc/src/hal_replacement.hpp +++ b/modules/imgproc/src/hal_replacement.hpp @@ -1572,6 +1572,32 @@ inline int hal_ni_calcHist(const uchar* src_data, size_t src_step, int src_type, #define cv_hal_calcHist hal_ni_calcHist //! @endcond +/** + @brief Compares a template against overlapped image regions. + @param src_data Source image (single-channel, CV_8U or CV_32F) data + @param src_step Source image step + @param src_width Source image width + @param src_height Source image height + @param templ_data Template image data (same type as source) + @param templ_step Template image step + @param templ_width Template image width + @param templ_height Template image height + @param result_data Destination map data (single-channel CV_32F, size (src_width-templ_width+1) x (src_height-templ_height+1)) + @param result_step Destination map step + @param depth Depth of source and template images (CV_8U or CV_32F) + @param cn Number of channels + @param method Comparison method (cv::TemplateMatchModes) + @sa matchTemplate +*/ +inline int hal_ni_matchTemplate(const uchar* src_data, size_t src_step, int src_width, int src_height, + const uchar* templ_data, size_t templ_step, int templ_width, int templ_height, + float* result_data, size_t result_step, int depth, int cn, int method) +{ return CV_HAL_ERROR_NOT_IMPLEMENTED; } + +//! @cond IGNORED +#define cv_hal_matchTemplate hal_ni_matchTemplate +//! @endcond + //! @} #if defined(__clang__) diff --git a/modules/imgproc/src/templmatch.cpp b/modules/imgproc/src/templmatch.cpp index 466823819e..6e25607b37 100644 --- a/modules/imgproc/src/templmatch.cpp +++ b/modules/imgproc/src/templmatch.cpp @@ -1034,135 +1034,6 @@ static void common_matchTemplate( Mat& img, Mat& templ, Mat& result, int method, } } - -#if defined HAVE_IPP -namespace cv -{ -typedef IppStatus (CV_STDCALL * ippimatchTemplate)(const void*, int, IppiSize, const void*, int, IppiSize, Ipp32f* , int , IppEnum , Ipp8u*); - -static bool ipp_crossCorr(const Mat& src, const Mat& tpl, Mat& dst, bool normed) -{ - CV_INSTRUMENT_REGION_IPP(); - - IppStatus status; - - IppiSize srcRoiSize = {src.cols,src.rows}; - IppiSize tplRoiSize = {tpl.cols,tpl.rows}; - - IppAutoBuffer buffer; - int bufSize=0; - - int depth = src.depth(); - - ippimatchTemplate ippiCrossCorrNorm = - depth==CV_8U ? (ippimatchTemplate)ippiCrossCorrNorm_8u32f_C1R: - depth==CV_32F? (ippimatchTemplate)ippiCrossCorrNorm_32f_C1R: 0; - - if (ippiCrossCorrNorm==0) - return false; - - IppEnum funCfg = (IppEnum)(+ippAlgAuto | ippiROIValid); - if(normed) - funCfg |= ippiNorm; - else - funCfg |= ippiNormNone; - - status = ippiCrossCorrNormGetBufferSize(srcRoiSize, tplRoiSize, funCfg, &bufSize); - if ( status < 0 ) - return false; - - buffer.allocate( bufSize ); - - status = CV_INSTRUMENT_FUN_IPP(ippiCrossCorrNorm, src.ptr(), (int)src.step, srcRoiSize, tpl.ptr(), (int)tpl.step, tplRoiSize, dst.ptr(), (int)dst.step, funCfg, buffer); - return status >= 0; -} - -static bool ipp_sqrDistance(const Mat& src, const Mat& tpl, Mat& dst) -{ - CV_INSTRUMENT_REGION_IPP(); - - IppStatus status; - - IppiSize srcRoiSize = {src.cols,src.rows}; - IppiSize tplRoiSize = {tpl.cols,tpl.rows}; - - IppAutoBuffer buffer; - int bufSize=0; - - int depth = src.depth(); - - ippimatchTemplate ippiSqrDistanceNorm = - depth==CV_8U ? (ippimatchTemplate)ippiSqrDistanceNorm_8u32f_C1R: - depth==CV_32F? (ippimatchTemplate)ippiSqrDistanceNorm_32f_C1R: 0; - - if (ippiSqrDistanceNorm==0) - return false; - - IppEnum funCfg = (IppEnum)(+ippAlgAuto | ippiROIValid | ippiNormNone); - status = ippiSqrDistanceNormGetBufferSize(srcRoiSize, tplRoiSize, funCfg, &bufSize); - if ( status < 0 ) - return false; - - buffer.allocate( bufSize ); - - status = CV_INSTRUMENT_FUN_IPP(ippiSqrDistanceNorm, src.ptr(), (int)src.step, srcRoiSize, tpl.ptr(), (int)tpl.step, tplRoiSize, dst.ptr(), (int)dst.step, funCfg, buffer); - dst = cv::max(dst, 0); // handle edge case from rounding in variance computation which can result in negative values - return status >= 0; -} - -static bool ipp_matchTemplate( Mat& img, Mat& templ, Mat& result, int method) -{ - CV_INSTRUMENT_REGION_IPP(); - - if(img.channels() != 1) - return false; - - // These functions are not efficient if template size is comparable with image size - if(templ.size().area()*4 > img.size().area()) - return false; - - // CV_8U SQDIFF/SQDIFF_NORMED suffer from float32 catastrophic cancellation - // in IPP's internal accumulators; fall through to the double-precision path instead. - if(img.depth() == CV_8U && (method == cv::TM_SQDIFF || method == cv::TM_SQDIFF_NORMED)) - return false; - - if(method == cv::TM_SQDIFF) - { - if(ipp_sqrDistance(img, templ, result)) - return true; - } - else if(method == cv::TM_SQDIFF_NORMED) - { - if(ipp_crossCorr(img, templ, result, false)) - { - common_matchTemplate(img, templ, result, cv::TM_SQDIFF_NORMED, 1); - return true; - } - } - else if(method == cv::TM_CCORR) - { - if(ipp_crossCorr(img, templ, result, false)) - return true; - } - else if(method == cv::TM_CCORR_NORMED) - { - if(ipp_crossCorr(img, templ, result, true)) - return true; - } - else if(method == cv::TM_CCOEFF || method == cv::TM_CCOEFF_NORMED) - { - if(ipp_crossCorr(img, templ, result, false)) - { - common_matchTemplate(img, templ, result, method, 1); - return true; - } - } - - return false; -} -} -#endif - //////////////////////////////////////////////////////////////////////////////////////////////////////// void cv::matchTemplate( InputArray _img, InputArray _templ, OutputArray _result, int method, InputArray _mask ) @@ -1196,7 +1067,22 @@ void cv::matchTemplate( InputArray _img, InputArray _templ, OutputArray _result, _result.create(corrSize, CV_32F); Mat result = _result.getMat(); - CV_IPP_RUN_FAST(ipp_matchTemplate(img, templ, result, method)) + { + int hal_res = cv_hal_matchTemplate(img.data, img.step, img.cols, img.rows, + templ.data, templ.step, templ.cols, templ.rows, + result.ptr(), result.step, depth, cn, method); + if (hal_res == CV_HAL_ERROR_OK) + { + if (method == cv::TM_SQDIFF_NORMED || method == cv::TM_CCOEFF || method == cv::TM_CCOEFF_NORMED) + common_matchTemplate(img, templ, result, method, 1); + return; + } + else if (hal_res != CV_HAL_ERROR_NOT_IMPLEMENTED) + { + CV_Error_(cv::Error::StsInternal, + ("HAL implementation matchTemplate ==> cv_hal_matchTemplate returned %d (0x%08x)", hal_res, hal_res)); + } + } bool use64f = (depth == CV_8U) && (method == cv::TM_SQDIFF || method == cv::TM_SQDIFF_NORMED); Mat result64f;