generic-library/opencv
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

refactored imgproc

Browse Source
This commit is contained in:
Ilya Lavrenov
2014-01-27 13:25:21 +04:00
parent 321782b9b7
commit 78c2b3ca2a
17 changed files with 479 additions and 432 deletions
Download Patch File Download Diff File Expand all files Collapse all files

617
modules/imgproc/src/templmatch.cpp
View File

@@ -42,338 +42,326 @@
#include "precomp.hpp"
#include "opencl_kernels.hpp"
//////////////////////////////////////////////////matchTemplate//////////////////////////////////////////////////////////
////////////////////////////////////////////////// matchTemplate //////////////////////////////////////////////////////////
namespace cv
{
static bool matchTemplate_CCORR(InputArray _image, InputArray _templ, OutputArray _result);
static bool matchTemplate_CCORR_NORMED(InputArray _image, InputArray _templ, OutputArray _result);
static bool matchTemplate_SQDIFF(InputArray _image, InputArray _templ, OutputArray _result);
static bool matchTemplate_SQDIFF_NORMED (InputArray _image, InputArray _templ, OutputArray _result);
#ifdef HAVE_OPENCL
static bool matchTemplate_CCOEFF(InputArray _image, InputArray _templ, OutputArray _result);
static bool matchTemplate_CCOEFF_NORMED(InputArray _image, InputArray _templ, OutputArray _result);
static bool matchTemplateNaive_CCORR (InputArray _image, InputArray _templ, OutputArray _result, int cn);
static bool matchTemplateNaive_SQDIFF(InputArray _image, InputArray _templ, OutputArray _result, int cn);
static bool useNaive(int method, int depth, Size size)
{
static bool useNaive(int method, int depth, Size size)
{
#ifdef HAVE_CLAMDFFT
if (method == TM_SQDIFF && depth == CV_32F)
return true;
else if(method == TM_CCORR || (method == TM_SQDIFF && depth == CV_8U))
return size.height < 18 && size.width < 18;
else
return false;
if (method == TM_SQDIFF && depth == CV_32F)
return true;
else if(method == TM_CCORR || (method == TM_SQDIFF && depth == CV_8U))
return size.height < 18 && size.width < 18;
else
return false;
#else
#define UNUSED(x) (void)(x);
UNUSED(method) UNUSED(depth) UNUSED(size)
UNUSED(method) UNUSED(depth) UNUSED(size)
#undef UNUSED
return true;
return true;
#endif
}
}
///////////////////////////////////////////////////CCORR//////////////////////////////////////////////////////////////
/////////////////////////////////////////////////// CCORR //////////////////////////////////////////////////////////////
static bool matchTemplate_CCORR(InputArray _image, InputArray _templ, OutputArray _result)
{
if (useNaive(TM_CCORR, _image.depth(), _templ.size()) )
return matchTemplateNaive_CCORR(_image, _templ, _result, _image.channels());
else
return false;
}
static bool matchTemplateNaive_CCORR (InputArray _image, InputArray _templ, OutputArray _result, int cn)
{
int type = _image.type();
int depth = CV_MAT_DEPTH(type);
static bool matchTemplateNaive_CCORR (InputArray _image, InputArray _templ, OutputArray _result, int cn)
const char * kernelName = "matchTemplate_Naive_CCORR";
ocl::Kernel k (kernelName, ocl::imgproc::match_template_oclsrc, format("-D type=%s -D elem_type=%s -D cn=%d",ocl::typeToStr(type), ocl::typeToStr(depth), cn));
if (k.empty())
return false;
UMat image = _image.getUMat();
UMat templ = _templ.getUMat(), result;
_result.create(image.rows - templ.rows + 1, image.cols - templ.cols + 1, CV_32F);
result = _result.getUMat();
size_t globalsize[2] = {result.cols, result.rows};
return k.args(ocl::KernelArg::ReadOnlyNoSize(image), ocl::KernelArg::ReadOnly(templ), ocl::KernelArg::WriteOnly(result)).run(2,globalsize,NULL,false);
}
static bool matchTemplate_CCORR_NORMED(InputArray _image, InputArray _templ, OutputArray _result)
{
matchTemplate(_image, _templ, _result, CV_TM_CCORR);
int type = _image.type();
int depth = CV_MAT_DEPTH(type), cn = CV_MAT_CN(type);
const char * kernelName = "matchTemplate_CCORR_NORMED";
ocl::Kernel k(kernelName, ocl::imgproc::match_template_oclsrc, format("-D type=%s -D elem_type=%s -D cn=%d",ocl::typeToStr(type), ocl::typeToStr(depth), cn));
if (k.empty())
return false;
UMat image = _image.getUMat();
UMat templ = _templ.getUMat(), result;
_result.create(image.rows - templ.rows + 1, image.cols - templ.cols + 1, CV_32F);
result = _result.getUMat();
UMat image_sums, image_sqsums;
integral(image.reshape(1), image_sums, image_sqsums, CV_32F, CV_32F);
UMat templ_resh, temp;
templ.reshape(1).convertTo(templ_resh, CV_32F);
multiply(templ_resh, templ_resh, temp);
unsigned long long templ_sqsum = (unsigned long long)sum(temp)[0];
size_t globalsize[2] = {result.cols, result.rows};
return k.args(ocl::KernelArg::ReadOnlyNoSize(image_sqsums), ocl::KernelArg::WriteOnly(result), templ.rows, templ.cols, templ_sqsum).run(2,globalsize,NULL,false);
}
static bool matchTemplate_CCORR(InputArray _image, InputArray _templ, OutputArray _result)
{
if (useNaive(TM_CCORR, _image.depth(), _templ.size()) )
return matchTemplateNaive_CCORR(_image, _templ, _result, _image.channels());
else
return false;
}
////////////////////////////////////// SQDIFF //////////////////////////////////////////////////////////////
static bool matchTemplateNaive_SQDIFF(InputArray _image, InputArray _templ, OutputArray _result, int cn)
{
int type = _image.type();
int depth = CV_MAT_DEPTH(type);
const char * kernelName = "matchTemplate_Naive_SQDIFF";
ocl::Kernel k (kernelName, ocl::imgproc::match_template_oclsrc, format("-D type=%s -D elem_type=%s -D cn=%d",ocl::typeToStr(type), ocl::typeToStr(depth), cn));
if (k.empty())
return false;
UMat image = _image.getUMat();
UMat templ = _templ.getUMat(), result;
_result.create(image.rows - templ.rows + 1, image.cols - templ.cols + 1, CV_32F);
result = _result.getUMat();
size_t globalsize[2] = {result.cols, result.rows};
return k.args(ocl::KernelArg::ReadOnlyNoSize(image), ocl::KernelArg::ReadOnly(templ), ocl::KernelArg::WriteOnly(result)).run(2,globalsize,NULL,false);
}
static bool matchTemplate_SQDIFF_NORMED (InputArray _image, InputArray _templ, OutputArray _result)
{
matchTemplate(_image, _templ, _result, CV_TM_CCORR);
int type = _image.type();
int depth = CV_MAT_DEPTH(type), cn = CV_MAT_CN(type);
const char * kernelName = "matchTemplate_SQDIFF_NORMED";
ocl::Kernel k(kernelName, ocl::imgproc::match_template_oclsrc, format("-D type=%s -D elem_type=%s -D cn=%d",ocl::typeToStr(type), ocl::typeToStr(depth), cn));
if (k.empty())
return false;
UMat image = _image.getUMat();
UMat templ = _templ.getUMat(), result;
_result.create(image.rows - templ.rows + 1, image.cols - templ.cols + 1, CV_32F);
result = _result.getUMat();
UMat image_sums, image_sqsums;
integral(image.reshape(1), image_sums, image_sqsums, CV_32F, CV_32F);
UMat temp, templ_resh;
templ.reshape(1).convertTo(templ_resh, CV_32F);
multiply(templ_resh, templ_resh, temp);
unsigned long long templ_sqsum = (unsigned long long)sum(temp)[0];
size_t globalsize[2] = {result.cols, result.rows};
return k.args(ocl::KernelArg::ReadOnlyNoSize(image_sqsums), ocl::KernelArg::WriteOnly(result), templ.rows, templ.cols, templ_sqsum).run(2,globalsize,NULL,false);
}
static bool matchTemplate_SQDIFF(InputArray _image, InputArray _templ, OutputArray _result)
{
if (useNaive(TM_SQDIFF, _image.depth(), _templ.size()))
return matchTemplateNaive_SQDIFF(_image, _templ, _result, _image.channels());
else
return false;
}
///////////////////////////////////// CCOEFF /////////////////////////////////////////////////////////////////
static bool matchTemplate_CCOEFF(InputArray _image, InputArray _templ, OutputArray _result)
{
matchTemplate(_image, _templ, _result, CV_TM_CCORR);
UMat image_sums;
integral(_image, image_sums);
int type = image_sums.type();
int depth = CV_MAT_DEPTH(type), cn = CV_MAT_CN(type);
const char * kernelName;
if (cn==1)
kernelName = "matchTemplate_Prepared_CCOEFF_C1";
else if (cn==2)
kernelName = "matchTemplate_Prepared_CCOEFF_C2";
else
kernelName = "matchTemplate_Prepared_CCOEFF_C4";
ocl::Kernel k(kernelName, ocl::imgproc::match_template_oclsrc, format("-D type=%s -D elem_type=%s -D cn=%d",ocl::typeToStr(type), ocl::typeToStr(depth), cn));
if (k.empty())
return false;
UMat templ = _templ.getUMat(), result;
Size size = _image.size();
_result.create(size.height - templ.rows + 1, size.width - templ.cols + 1, CV_32F);
result = _result.getUMat();
size_t globalsize[2] = {result.cols, result.rows};
if (cn==1)
{
int type = _image.type();
int depth = CV_MAT_DEPTH(type);
const char * kernelName = "matchTemplate_Naive_CCORR";
ocl::Kernel k (kernelName, ocl::imgproc::match_template_oclsrc, format("-D type=%s -D elem_type=%s -D cn=%d",ocl::typeToStr(type), ocl::typeToStr(depth), cn));
if (k.empty())
return false;
UMat image = _image.getUMat();
UMat templ = _templ.getUMat(), result;
_result.create(image.rows - templ.rows + 1, image.cols - templ.cols + 1, CV_32F);
result = _result.getUMat();
size_t globalsize[2] = {result.cols, result.rows};
return k.args(ocl::KernelArg::ReadOnlyNoSize(image), ocl::KernelArg::ReadOnly(templ), ocl::KernelArg::WriteOnly(result)).run(2,globalsize,NULL,false);
float templ_sum = (float)sum(_templ)[0]/ _templ.size().area();
return k.args(ocl::KernelArg::ReadOnlyNoSize(image_sums), ocl::KernelArg::WriteOnly(result), templ.rows, templ.cols, templ_sum).run(2,globalsize,NULL,false);
}
static bool matchTemplate_CCORR_NORMED(InputArray _image, InputArray _templ, OutputArray _result)
else
{
matchTemplate(_image, _templ, _result, CV_TM_CCORR);
int type = _image.type();
int depth = CV_MAT_DEPTH(type), cn = CV_MAT_CN(type);
const char * kernelName = "matchTemplate_CCORR_NORMED";
ocl::Kernel k(kernelName, ocl::imgproc::match_template_oclsrc, format("-D type=%s -D elem_type=%s -D cn=%d",ocl::typeToStr(type), ocl::typeToStr(depth), cn));
if (k.empty())
return false;
UMat image = _image.getUMat();
UMat templ = _templ.getUMat(), result;
_result.create(image.rows - templ.rows + 1, image.cols - templ.cols + 1, CV_32F);
result = _result.getUMat();
UMat image_sums, image_sqsums;
integral(image.reshape(1), image_sums, image_sqsums, CV_32F, CV_32F);
UMat templ_resh, temp;
templ.reshape(1).convertTo(templ_resh, CV_32F);
multiply(templ_resh, templ_resh, temp);
unsigned long long templ_sqsum = (unsigned long long)sum(temp)[0];
size_t globalsize[2] = {result.cols, result.rows};
return k.args(ocl::KernelArg::ReadOnlyNoSize(image_sqsums), ocl::KernelArg::WriteOnly(result), templ.rows, templ.cols, templ_sqsum).run(2,globalsize,NULL,false);
}
//////////////////////////////////////SQDIFF//////////////////////////////////////////////////////////////
static bool matchTemplate_SQDIFF(InputArray _image, InputArray _templ, OutputArray _result)
{
if (useNaive(TM_SQDIFF, _image.depth(), _templ.size()))
{
return matchTemplateNaive_SQDIFF(_image, _templ, _result, _image.channels());;
}
else
return false;
}
static bool matchTemplateNaive_SQDIFF(InputArray _image, InputArray _templ, OutputArray _result, int cn)
{
int type = _image.type();
int depth = CV_MAT_DEPTH(type);
const char * kernelName = "matchTemplate_Naive_SQDIFF";
ocl::Kernel k (kernelName, ocl::imgproc::match_template_oclsrc, format("-D type=%s -D elem_type=%s -D cn=%d",ocl::typeToStr(type), ocl::typeToStr(depth), cn));
if (k.empty())
return false;
UMat image = _image.getUMat();
UMat templ = _templ.getUMat(), result;
_result.create(image.rows - templ.rows + 1, image.cols - templ.cols + 1, CV_32F);
result = _result.getUMat();
size_t globalsize[2] = {result.cols, result.rows};
return k.args(ocl::KernelArg::ReadOnlyNoSize(image), ocl::KernelArg::ReadOnly(templ), ocl::KernelArg::WriteOnly(result)).run(2,globalsize,NULL,false);
}
static bool matchTemplate_SQDIFF_NORMED (InputArray _image, InputArray _templ, OutputArray _result)
{
matchTemplate(_image, _templ, _result, CV_TM_CCORR);
int type = _image.type();
int depth = CV_MAT_DEPTH(type), cn = CV_MAT_CN(type);
const char * kernelName = "matchTemplate_SQDIFF_NORMED";
ocl::Kernel k(kernelName, ocl::imgproc::match_template_oclsrc, format("-D type=%s -D elem_type=%s -D cn=%d",ocl::typeToStr(type), ocl::typeToStr(depth), cn));
if (k.empty())
return false;
UMat image = _image.getUMat();
UMat templ = _templ.getUMat(), result;
_result.create(image.rows - templ.rows + 1, image.cols - templ.cols + 1, CV_32F);
result = _result.getUMat();
UMat image_sums, image_sqsums;
integral(image.reshape(1), image_sums, image_sqsums, CV_32F, CV_32F);
UMat temp, templ_resh;
templ.reshape(1).convertTo(templ_resh, CV_32F);
multiply(templ_resh, templ_resh, temp);
unsigned long long templ_sqsum = (unsigned long long)sum(temp)[0];
size_t globalsize[2] = {result.cols, result.rows};
return k.args(ocl::KernelArg::ReadOnlyNoSize(image_sqsums), ocl::KernelArg::WriteOnly(result), templ.rows, templ.cols, templ_sqsum).run(2,globalsize,NULL,false);
}
/////////////////////////////////////CCOEFF/////////////////////////////////////////////////////////////////
static bool matchTemplate_CCOEFF(InputArray _image, InputArray _templ, OutputArray _result)
{
matchTemplate(_image, _templ, _result, CV_TM_CCORR);
UMat image_sums;
integral(_image, image_sums);
int type = image_sums.type();
int depth = CV_MAT_DEPTH(type), cn = CV_MAT_CN(type);
const char * kernelName;
if (cn==1)
kernelName = "matchTemplate_Prepared_CCOEFF_C1";
else if (cn==2)
kernelName = "matchTemplate_Prepared_CCOEFF_C2";
else
kernelName = "matchTemplate_Prepared_CCOEFF_C4";
ocl::Kernel k(kernelName, ocl::imgproc::match_template_oclsrc, format("-D type=%s -D elem_type=%s -D cn=%d",ocl::typeToStr(type), ocl::typeToStr(depth), cn));
if (k.empty())
return false;
UMat templ = _templ.getUMat(), result;
Size size = _image.size();
_result.create(size.height - templ.rows + 1, size.width - templ.cols + 1, CV_32F);
result = _result.getUMat();
size_t globalsize[2] = {result.cols, result.rows};
if (cn==1)
{
float templ_sum = (float)sum(_templ)[0]/ _templ.size().area();
return k.args(ocl::KernelArg::ReadOnlyNoSize(image_sums), ocl::KernelArg::WriteOnly(result), templ.rows, templ.cols, templ_sum).run(2,globalsize,NULL,false);
}
else
{
Vec4f templ_sum = Vec4f::all(0);
templ_sum = sum(templ)/ templ.size().area();
if (cn==2)
return k.args(ocl::KernelArg::ReadOnlyNoSize(image_sums), ocl::KernelArg::WriteOnly(result), templ.rows, templ.cols,
templ_sum[0],templ_sum[1]).run(2,globalsize,NULL,false);
Vec4f templ_sum = Vec4f::all(0);
templ_sum = sum(templ)/ templ.size().area();
if (cn==2)
return k.args(ocl::KernelArg::ReadOnlyNoSize(image_sums), ocl::KernelArg::WriteOnly(result), templ.rows, templ.cols,
templ_sum[0],templ_sum[1],templ_sum[2],templ_sum[3]).run(2,globalsize,NULL,false);
}
}
templ_sum[0],templ_sum[1]).run(2,globalsize,NULL,false);
static bool matchTemplate_CCOEFF_NORMED(InputArray _image, InputArray _templ, OutputArray _result)
{
UMat imagef, templf;
_image.getUMat().convertTo(imagef, CV_32F);
_templ.getUMat().convertTo(templf, CV_32F);
matchTemplate(imagef, templf, _result, CV_TM_CCORR);
const char * kernelName;
UMat temp, image_sums, image_sqsums;
integral(_image,image_sums, image_sqsums, CV_32F, CV_32F);
int type = image_sums.type();
int depth = CV_MAT_DEPTH(type), cn = CV_MAT_CN(type);
if (cn== 1)
kernelName = "matchTemplate_CCOEFF_NORMED_C1";
else if (cn==2)
kernelName = "matchTemplate_CCOEFF_NORMED_C2";
else
kernelName = "matchTemplate_CCOEFF_NORMED_C4";
ocl::Kernel k(kernelName, ocl::imgproc::match_template_oclsrc,
format("-D type=%s -D elem_type=%s -D cn=%d", ocl::typeToStr(type), ocl::typeToStr(depth), cn));
if (k.empty())
return false;
UMat image = _image.getUMat();
UMat templ = _templ.getUMat(), result;
int image_rows = _image.size().height, image_cols = _image.size().width;
_result.create(image_rows - templ.rows + 1, image_cols - templ.cols + 1, CV_32F);
result = _result.getUMat();
size_t globalsize[2] = {result.cols, result.rows};
float scale = 1.f / templ.size().area();
if (cn==1)
{
float templ_sum = (float)sum(templ)[0];
multiply(templf, templf, temp);
float templ_sqsum = (float)sum(temp)[0];
templ_sqsum -= scale * templ_sum * templ_sum;
templ_sum *= scale;
if (templ_sqsum < DBL_EPSILON)
{
result = Scalar::all(1);
return true;
}
return k.args(ocl::KernelArg::ReadOnlyNoSize(image_sums),ocl::KernelArg::ReadOnlyNoSize(image_sqsums),
ocl::KernelArg::WriteOnly(result), templ.rows, templ.cols, scale, templ_sum, templ_sqsum)
.run(2,globalsize,NULL,false);
}
else
{
Vec4f templ_sum = Vec4f::all(0);
Vec4f templ_sqsum = Vec4f::all(0);
templ_sum = sum(templ);
multiply(templf, templf, temp);
templ_sqsum = sum(temp);
float templ_sqsum_sum = 0;
for(int i = 0; i < cn; i ++)
{
templ_sqsum_sum += templ_sqsum[i] - scale * templ_sum[i] * templ_sum[i];
}
templ_sum *= scale;
if (templ_sqsum_sum < DBL_EPSILON)
{
result = Scalar::all(1);
return true;
}
if (cn==2)
return k.args(ocl::KernelArg::ReadOnlyNoSize(image_sums), ocl::KernelArg::ReadOnlyNoSize(image_sqsums),
ocl::KernelArg::WriteOnly(result), templ.rows, templ.cols, scale,
templ_sum[0],templ_sum[1], templ_sqsum_sum)
.run(2,globalsize,NULL,false);
return k.args(ocl::KernelArg::ReadOnlyNoSize(image_sums), ocl::KernelArg::ReadOnlyNoSize(image_sqsums),
ocl::KernelArg::WriteOnly(result), templ.rows, templ.cols, scale,
templ_sum[0],templ_sum[1],templ_sum[2],templ_sum[3], templ_sqsum_sum)
.run(2,globalsize,NULL,false);
}
}
///////////////////////////////////////////////////////////////////////////////////////////////////////////
static bool ocl_matchTemplate( InputArray _img, InputArray _templ, OutputArray _result, int method)
{
int cn = CV_MAT_CN(_img.type());
if (cn == 3 || cn > 4)
return false;
typedef bool (*Caller)(InputArray _img, InputArray _templ, OutputArray _result);
const Caller callers[] =
{
matchTemplate_SQDIFF, matchTemplate_SQDIFF_NORMED, matchTemplate_CCORR,
matchTemplate_CCORR_NORMED, matchTemplate_CCOEFF, matchTemplate_CCOEFF_NORMED
};
Caller caller = callers[method];
return caller(_img, _templ, _result);
return k.args(ocl::KernelArg::ReadOnlyNoSize(image_sums), ocl::KernelArg::WriteOnly(result), templ.rows, templ.cols,
templ_sum[0],templ_sum[1],templ_sum[2],templ_sum[3]).run(2,globalsize,NULL,false);
}
}
namespace cv
static bool matchTemplate_CCOEFF_NORMED(InputArray _image, InputArray _templ, OutputArray _result)
{
UMat imagef, templf;
_image.getUMat().convertTo(imagef, CV_32F);
_templ.getUMat().convertTo(templf, CV_32F);
matchTemplate(imagef, templf, _result, CV_TM_CCORR);
const char * kernelName;
UMat temp, image_sums, image_sqsums;
integral(_image,image_sums, image_sqsums, CV_32F, CV_32F);
int type = image_sums.type();
int depth = CV_MAT_DEPTH(type), cn = CV_MAT_CN(type);
if (cn== 1)
kernelName = "matchTemplate_CCOEFF_NORMED_C1";
else if (cn==2)
kernelName = "matchTemplate_CCOEFF_NORMED_C2";
else
kernelName = "matchTemplate_CCOEFF_NORMED_C4";
ocl::Kernel k(kernelName, ocl::imgproc::match_template_oclsrc,
format("-D type=%s -D elem_type=%s -D cn=%d", ocl::typeToStr(type), ocl::typeToStr(depth), cn));
if (k.empty())
return false;
UMat image = _image.getUMat();
UMat templ = _templ.getUMat(), result;
int image_rows = _image.size().height, image_cols = _image.size().width;
_result.create(image_rows - templ.rows + 1, image_cols - templ.cols + 1, CV_32F);
result = _result.getUMat();
size_t globalsize[2] = {result.cols, result.rows};
float scale = 1.f / templ.size().area();
if (cn==1)
{
float templ_sum = (float)sum(templ)[0];
multiply(templf, templf, temp);
float templ_sqsum = (float)sum(temp)[0];
templ_sqsum -= scale * templ_sum * templ_sum;
templ_sum *= scale;
if (templ_sqsum < DBL_EPSILON)
{
result = Scalar::all(1);
return true;
}
return k.args(ocl::KernelArg::ReadOnlyNoSize(image_sums),ocl::KernelArg::ReadOnlyNoSize(image_sqsums),
ocl::KernelArg::WriteOnly(result), templ.rows, templ.cols, scale, templ_sum, templ_sqsum)
.run(2,globalsize,NULL,false);
}
else
{
Vec4f templ_sum = Vec4f::all(0);
Vec4f templ_sqsum = Vec4f::all(0);
templ_sum = sum(templ);
multiply(templf, templf, temp);
templ_sqsum = sum(temp);
float templ_sqsum_sum = 0;
for(int i = 0; i < cn; i ++)
{
templ_sqsum_sum += templ_sqsum[i] - scale * templ_sum[i] * templ_sum[i];
}
templ_sum *= scale;
if (templ_sqsum_sum < DBL_EPSILON)
{
result = Scalar::all(1);
return true;
}
if (cn==2)
return k.args(ocl::KernelArg::ReadOnlyNoSize(image_sums), ocl::KernelArg::ReadOnlyNoSize(image_sqsums),
ocl::KernelArg::WriteOnly(result), templ.rows, templ.cols, scale,
templ_sum[0],templ_sum[1], templ_sqsum_sum)
.run(2,globalsize,NULL,false);
return k.args(ocl::KernelArg::ReadOnlyNoSize(image_sums), ocl::KernelArg::ReadOnlyNoSize(image_sqsums),
ocl::KernelArg::WriteOnly(result), templ.rows, templ.cols, scale,
templ_sum[0],templ_sum[1],templ_sum[2],templ_sum[3], templ_sqsum_sum)
.run(2,globalsize,NULL,false);
}
}
///////////////////////////////////////////////////////////////////////////////////////////////////////////
static bool ocl_matchTemplate( InputArray _img, InputArray _templ, OutputArray _result, int method)
{
int cn = CV_MAT_CN(_img.type());
if (cn == 3 || cn > 4)
return false;
typedef bool (*Caller)(InputArray _img, InputArray _templ, OutputArray _result);
const Caller callers[] =
{
matchTemplate_SQDIFF, matchTemplate_SQDIFF_NORMED, matchTemplate_CCORR,
matchTemplate_CCORR_NORMED, matchTemplate_CCOEFF, matchTemplate_CCOEFF_NORMED
};
Caller caller = callers[method];
return caller(_img, _templ, _result);
}
#endif
void crossCorr( const Mat& img, const Mat& _templ, Mat& corr,
Size corrsize, int ctype,
@@ -564,9 +552,7 @@ void crossCorr( const Mat& img, const Mat& _templ, Mat& corr,
void cv::matchTemplate( InputArray _img, InputArray _templ, OutputArray _result, int method )
{
CV_Assert( CV_TM_SQDIFF <= method && method <= CV_TM_CCOEFF_NORMED );
CV_Assert( (_img.depth() == CV_8U || _img.depth() == CV_32F) && _img.type() == _templ.type() );
CV_Assert(_img.dims() <= 2);
bool swapNotNeed = (_img.size().height >= _templ.size().height && _img.size().width >= _templ.size().width);
@@ -575,9 +561,8 @@ void cv::matchTemplate( InputArray _img, InputArray _templ, OutputArray _result,
CV_Assert(_img.size().height <= _templ.size().height && _img.size().width <= _templ.size().width);
}
bool use_opencl = ocl::useOpenCL() && _result.isUMat();
if ( use_opencl && (swapNotNeed ? ocl_matchTemplate(_img,_templ,_result,method) : ocl_matchTemplate(_templ,_img,_result,method)))
return;
CV_OCL_RUN(_img.dims() <= 2 && _result.isUMat(),
(swapNotNeed ? ocl_matchTemplate(_img,_templ,_result,method) : ocl_matchTemplate(_templ,_img,_result,method)))
int numType = method == CV_TM_CCORR || method == CV_TM_CCORR_NORMED ? 0 :
method == CV_TM_CCOEFF || method == CV_TM_CCOEFF_NORMED ? 1 : 2;