CV_IMPL void
cvMatchTemplate( const CvArr* _img, const CvArr* _templ, CvArr* _result, int method )
{
CvMat* sum = 0;
CvMat* sqsum = 0;
CV_FUNCNAME( "cvMatchTemplate" );
__BEGIN__;
int coi1 = 0, coi2 = 0;
int depth, cn;
int i, j, k;
CvMat stub, *img = (CvMat*)_img;
CvMat tstub, *templ = (CvMat*)_templ;
CvMat rstub, *result = (CvMat*)_result;
CvScalar templ_mean = cvScalarAll(0);
double templ_norm = 0, templ_sum2 = 0;
int idx = 0, idx2 = 0;
double *p0, *p1, *p2, *p3;
double *q0, *q1, *q2, *q3;
double inv_area;
int sum_step, sqsum_step;
int num_type = method == CV_TM_CCORR || method == CV_TM_CCORR_NORMED ? 0 :
method == CV_TM_CCOEFF || method == CV_TM_CCOEFF_NORMED ? 1 : 2;
int is_normed = method == CV_TM_CCORR_NORMED ||
method == CV_TM_SQDIFF_NORMED ||
method == CV_TM_CCOEFF_NORMED;
CV_CALL( img = cvGetMat( img, &stub, &coi1 ));
CV_CALL( templ = cvGetMat( templ, &tstub, &coi2 ));
CV_CALL( result = cvGetMat( result, &rstub ));
if( CV_MAT_DEPTH( img->type ) != CV_8U &&
CV_MAT_DEPTH( img->type ) != CV_32F )
CV_ERROR( CV_StsUnsupportedFormat,
"The function supports only 8u and 32f data types" );
if( !CV_ARE_TYPES_EQ( img, templ ))
CV_ERROR( CV_StsUnmatchedSizes, "image and template should have the same type" );
if( CV_MAT_TYPE( result->type ) != CV_32FC1 )
CV_ERROR( CV_StsUnsupportedFormat, "output image should have 32f type" );
if( img->rows < templ->rows || img->cols < templ->cols )
{
CvMat* t;
CV_SWAP( img, templ, t );
}
if( result->rows != img->rows - templ->rows + 1 ||
result->cols != img->cols - templ->cols + 1 )
CV_ERROR( CV_StsUnmatchedSizes, "output image should be (W - w + 1)x(H - h + 1)" );
if( method < CV_TM_SQDIFF || method > CV_TM_CCOEFF_NORMED )
CV_ERROR( CV_StsBadArg, "unknown comparison method" );
depth = CV_MAT_DEPTH(img->type);
cn = CV_MAT_CN(img->type);
if( is_normed && cn == 1 && templ->rows > 8 && templ->cols > 8 &&
img->rows > templ->cols && img->cols > templ->cols )
{
CvTemplMatchIPPFunc ipp_func =
depth == CV_8U ?
(method == CV_TM_SQDIFF_NORMED ? (CvTemplMatchIPPFunc)icvSqrDistanceValid_Norm_8u32f_C1R_p :
method == CV_TM_CCORR_NORMED ? (CvTemplMatchIPPFunc)icvCrossCorrValid_Norm_8u32f_C1R_p :
(CvTemplMatchIPPFunc)icvCrossCorrValid_NormLevel_8u32f_C1R_p) :
(method == CV_TM_SQDIFF_NORMED ? (CvTemplMatchIPPFunc)icvSqrDistanceValid_Norm_32f_C1R_p :
method == CV_TM_CCORR_NORMED ? (CvTemplMatchIPPFunc)icvCrossCorrValid_Norm_32f_C1R_p :
(CvTemplMatchIPPFunc)icvCrossCorrValid_NormLevel_32f_C1R_p);
if( ipp_func )
{
CvSize img_size = cvGetMatSize(img), templ_size = cvGetMatSize(templ);
IPPI_CALL( ipp_func( img->data.ptr, img->step ? img->step : CV_STUB_STEP,
img_size, templ->data.ptr,
templ->step ? templ->step : CV_STUB_STEP,
templ_size, result->data.ptr,
result->step ? result->step : CV_STUB_STEP ));
for( i = 0; i < result->rows; i++ )
{
float* rrow = (float*)(result->data.ptr + i*result->step);
for( j = 0; j < result->cols; j++ )
{
if( fabs(rrow[j]) > 1. )
rrow[j] = rrow[j] < 0 ? -1.f : 1.f;
}
}
EXIT;
}
}
CV_CALL( icvCrossCorr( img, templ, result ));
if( method == CV_TM_CCORR )
EXIT;
inv_area = 1./((double)templ->rows * templ->cols);
CV_CALL( sum = cvCreateMat( img->rows + 1, img->cols + 1,
CV_MAKETYPE( CV_64F, cn )));
if( method == CV_TM_CCOEFF )
{
CV_CALL( cvIntegral( img, sum, 0, 0 ));
CV_CALL( templ_mean = cvAvg( templ ));
q0 = q1 = q2 = q3 = 0;
}
else
{
CvScalar _templ_sdv = cvScalarAll(0);
CV_CALL( sqsum = cvCreateMat( img->rows + 1, img->cols + 1,
CV_MAKETYPE( CV_64F, cn )));
CV_CALL( cvIntegral( img, sum, sqsum, 0 ));
CV_CALL( cvAvgSdv( templ, &templ_mean, &_templ_sdv ));
templ_norm = CV_SQR(_templ_sdv.val[0]) + CV_SQR(_templ_sdv.val[1]) +
CV_SQR(_templ_sdv.val[2]) + CV_SQR(_templ_sdv.val[3]);
if( templ_norm < DBL_EPSILON && method == CV_TM_CCOEFF_NORMED )
{
cvSet( result, cvScalarAll(1.) );
EXIT;
}
templ_sum2 = templ_norm +
CV_SQR(templ_mean.val[0]) + CV_SQR(templ_mean.val[1]) +
CV_SQR(templ_mean.val[2]) + CV_SQR(templ_mean.val[3]);
if( num_type != 1 )
{
templ_mean = cvScalarAll(0);
templ_norm = templ_sum2;
}
templ_sum2 /= inv_area;
templ_norm = sqrt(templ_norm);
templ_norm /= sqrt(inv_area); // care of accuracy here
q0 = (double*)sqsum->data.ptr;
q1 = q0 + templ->cols*cn;
q2 = (double*)(sqsum->data.ptr + templ->rows*sqsum->step);
q3 = q2 + templ->cols*cn;
}
p0 = (double*)sum->data.ptr;
p1 = p0 + templ->cols*cn;
p2 = (double*)(sum->data.ptr + templ->rows*sum->step);
p3 = p2 + templ->cols*cn;
sum_step = sum ? sum->step / sizeof(double) : 0;
sqsum_step = sqsum ? sqsum->step / sizeof(double) : 0;
for( i = 0; i < result->rows; i++ )
{
float* rrow = (float*)(result->data.ptr + i*result->step);
idx = i * sum_step;
idx2 = i * sqsum_step;
for( j = 0; j < result->cols; j++, idx += cn, idx2 += cn )
{
double num = rrow[j], t;
double wnd_mean2 = 0, wnd_sum2 = 0;
if( num_type == 1 )
{
for( k = 0; k < cn; k++ )
{
t = p0[idx+k] - p1[idx+k] - p2[idx+k] + p3[idx+k];
wnd_mean2 += CV_SQR(t);
num -= t*templ_mean.val[k];
}
wnd_mean2 *= inv_area;
}
if( is_normed || num_type == 2 )
{
for( k = 0; k < cn; k++ )
{
t = q0[idx2+k] - q1[idx2+k] - q2[idx2+k] + q3[idx2+k];
wnd_sum2 += t;
}
if( num_type == 2 )
num = wnd_sum2 - 2*num + templ_sum2;
}
if( is_normed )
{
t = sqrt(MAX(wnd_sum2 - wnd_mean2,0))*templ_norm;
if( t > DBL_EPSILON )
{
num /= t;
if( fabs(num) > 1. )
num = num > 0 ? 1 : -1;
}
else
num = method != CV_TM_SQDIFF_NORMED || num < DBL_EPSILON ? 0 : 1;
}
rrow[j] = (float)num;
}
}
__END__;
cvReleaseMat( &sum );
cvReleaseMat( &sqsum );
}
CV_IMPL void
cvMatchTemplate( const CvArr* _img, const CvArr* _templ, CvArr* _result, int method )
{
cv::Ptr<CvMat> sum, sqsum;
int coi1 = 0, coi2 = 0;
int depth, cn;
int i, j, k;
CvMat stub, *img = (CvMat*)_img;
CvMat tstub, *templ = (CvMat*)_templ;
CvMat rstub, *result = (CvMat*)_result;
CvScalar templ_mean = cvScalarAll(0);
double templ_norm = 0, templ_sum2 = 0;
int idx = 0, idx2 = 0;
double *p0, *p1, *p2, *p3;
double *q0, *q1, *q2, *q3;
double inv_area;
int sum_step, sqsum_step;
int num_type = method == CV_TM_CCORR || method == CV_TM_CCORR_NORMED ? 0 :
method == CV_TM_CCOEFF || method == CV_TM_CCOEFF_NORMED ? 1 : 2;
int is_normed = method == CV_TM_CCORR_NORMED ||
method == CV_TM_SQDIFF_NORMED ||
method == CV_TM_CCOEFF_NORMED;
img = cvGetMat( img, &stub, &coi1 );
templ = cvGetMat( templ, &tstub, &coi2 );
result = cvGetMat( result, &rstub );
if( CV_MAT_DEPTH( img->type ) != CV_8U &&
CV_MAT_DEPTH( img->type ) != CV_32F )
CV_Error( CV_StsUnsupportedFormat,
"The function supports only 8u and 32f data types" );
if( !CV_ARE_TYPES_EQ( img, templ ))
CV_Error( CV_StsUnmatchedSizes, "image and template should have the same type" );
if( CV_MAT_TYPE( result->type ) != CV_32FC1 )
CV_Error( CV_StsUnsupportedFormat, "output image should have 32f type" );
if( img->rows < templ->rows || img->cols < templ->cols )
{
CvMat* t;
CV_SWAP( img, templ, t );
}
if( result->rows != img->rows - templ->rows + 1 ||
result->cols != img->cols - templ->cols + 1 )
CV_Error( CV_StsUnmatchedSizes, "output image should be (W - w + 1)x(H - h + 1)" );
if( method < CV_TM_SQDIFF || method > CV_TM_CCOEFF_NORMED )
CV_Error( CV_StsBadArg, "unknown comparison method" );
depth = CV_MAT_DEPTH(img->type);
cn = CV_MAT_CN(img->type);
icvCrossCorr( img, templ, result );
if( method == CV_TM_CCORR )
return;
inv_area = 1./((double)templ->rows * templ->cols);
sum = cvCreateMat( img->rows + 1, img->cols + 1, CV_MAKETYPE( CV_64F, cn ));
if( method == CV_TM_CCOEFF )
{
cvIntegral( img, sum, 0, 0 );
templ_mean = cvAvg( templ );
q0 = q1 = q2 = q3 = 0;
}
else
{
CvScalar _templ_sdv = cvScalarAll(0);
sqsum = cvCreateMat( img->rows + 1, img->cols + 1, CV_MAKETYPE( CV_64F, cn ));
cvIntegral( img, sum, sqsum, 0 );
cvAvgSdv( templ, &templ_mean, &_templ_sdv );
templ_norm = CV_SQR(_templ_sdv.val[0]) + CV_SQR(_templ_sdv.val[1]) +
CV_SQR(_templ_sdv.val[2]) + CV_SQR(_templ_sdv.val[3]);
if( templ_norm < DBL_EPSILON && method == CV_TM_CCOEFF_NORMED )
{
cvSet( result, cvScalarAll(1.) );
return;
}
templ_sum2 = templ_norm +
CV_SQR(templ_mean.val[0]) + CV_SQR(templ_mean.val[1]) +
CV_SQR(templ_mean.val[2]) + CV_SQR(templ_mean.val[3]);
if( num_type != 1 )
{
templ_mean = cvScalarAll(0);
templ_norm = templ_sum2;
}
templ_sum2 /= inv_area;
templ_norm = sqrt(templ_norm);
templ_norm /= sqrt(inv_area); // care of accuracy here
q0 = (double*)sqsum->data.ptr;
q1 = q0 + templ->cols*cn;
q2 = (double*)(sqsum->data.ptr + templ->rows*sqsum->step);
q3 = q2 + templ->cols*cn;
}
p0 = (double*)sum->data.ptr;
p1 = p0 + templ->cols*cn;
p2 = (double*)(sum->data.ptr + templ->rows*sum->step);
p3 = p2 + templ->cols*cn;
sum_step = sum ? sum->step / sizeof(double) : 0;
sqsum_step = sqsum ? sqsum->step / sizeof(double) : 0;
for( i = 0; i < result->rows; i++ )
{
float* rrow = (float*)(result->data.ptr + i*result->step);
idx = i * sum_step;
idx2 = i * sqsum_step;
for( j = 0; j < result->cols; j++, idx += cn, idx2 += cn )
{
double num = rrow[j], t;
double wnd_mean2 = 0, wnd_sum2 = 0;
if( num_type == 1 )
{
for( k = 0; k < cn; k++ )
{
t = p0[idx+k] - p1[idx+k] - p2[idx+k] + p3[idx+k];
wnd_mean2 += CV_SQR(t);
num -= t*templ_mean.val[k];
}
wnd_mean2 *= inv_area;
}
if( is_normed || num_type == 2 )
{
for( k = 0; k < cn; k++ )
{
t = q0[idx2+k] - q1[idx2+k] - q2[idx2+k] + q3[idx2+k];
wnd_sum2 += t;
}
if( num_type == 2 )
num = wnd_sum2 - 2*num + templ_sum2;
}
if( is_normed )
{
t = sqrt(MAX(wnd_sum2 - wnd_mean2,0))*templ_norm;
if( fabs(num) < t )
num /= t;
else if( fabs(num) < t*1.125 )
num = num > 0 ? 1 : -1;
else
num = method != CV_TM_SQDIFF_NORMED ? 0 : 1;
}
rrow[j] = (float)num;
}
}
}
void cv::matchTemplate( InputArray _img, InputArray _templ, OutputArray _result, int method )
{
CV_Assert( CV_TM_SQDIFF <= method && method <= CV_TM_CCOEFF_NORMED );
int numType = method == CV_TM_CCORR || method == CV_TM_CCORR_NORMED ? 0 :
method == CV_TM_CCOEFF || method == CV_TM_CCOEFF_NORMED ? 1 : 2;
bool isNormed = method == CV_TM_CCORR_NORMED ||
method == CV_TM_SQDIFF_NORMED ||
method == CV_TM_CCOEFF_NORMED;
Mat img = _img.getMat(), templ = _templ.getMat();
if( img.rows < templ.rows || img.cols < templ.cols )
std::swap(img, templ);
CV_Assert( (img.depth() == CV_8U || img.depth() == CV_32F) &&
img.type() == templ.type() );
Size corrSize(img.cols - templ.cols + 1, img.rows - templ.rows + 1);
_result.create(corrSize, CV_32F);
Mat result = _result.getMat();
int cn = img.channels();
crossCorr( img, templ, result, result.size(), result.type(), Point(0,0), 0, 0);
if( method == CV_TM_CCORR )
return;
double invArea = 1./((double)templ.rows * templ.cols);
Mat sum, sqsum;
Scalar templMean, templSdv;
double *q0 = 0, *q1 = 0, *q2 = 0, *q3 = 0;
double templNorm = 0, templSum2 = 0;
if( method == CV_TM_CCOEFF )
{
integral(img, sum, CV_64F);
templMean = mean(templ);
}
else
{
integral(img, sum, sqsum, CV_64F);
meanStdDev( templ, templMean, templSdv );
templNorm = CV_SQR(templSdv[0]) + CV_SQR(templSdv[1]) +
CV_SQR(templSdv[2]) + CV_SQR(templSdv[3]);
if( templNorm < DBL_EPSILON && method == CV_TM_CCOEFF_NORMED )
{
result = Scalar::all(1);
return;
}
templSum2 = templNorm +
CV_SQR(templMean[0]) + CV_SQR(templMean[1]) +
CV_SQR(templMean[2]) + CV_SQR(templMean[3]);
if( numType != 1 )
{
templMean = Scalar::all(0);
templNorm = templSum2;
}
templSum2 /= invArea;
templNorm = sqrt(templNorm);
templNorm /= sqrt(invArea); // care of accuracy here
q0 = (double*)sqsum.data;
q1 = q0 + templ.cols*cn;
q2 = (double*)(sqsum.data + templ.rows*sqsum.step);
q3 = q2 + templ.cols*cn;
}
double* p0 = (double*)sum.data;
double* p1 = p0 + templ.cols*cn;
double* p2 = (double*)(sum.data + templ.rows*sum.step);
double* p3 = p2 + templ.cols*cn;
int sumstep = sum.data ? (int)(sum.step / sizeof(double)) : 0;
int sqstep = sqsum.data ? (int)(sqsum.step / sizeof(double)) : 0;
int i, j, k;
for( i = 0; i < result.rows; i++ )
{
float* rrow = (float*)(result.data + i*result.step);
int idx = i * sumstep;
int idx2 = i * sqstep;
for( j = 0; j < result.cols; j++, idx += cn, idx2 += cn )
{
double num = rrow[j], t;
double wndMean2 = 0, wndSum2 = 0;
if( numType == 1 )
{
for( k = 0; k < cn; k++ )
{
t = p0[idx+k] - p1[idx+k] - p2[idx+k] + p3[idx+k];
wndMean2 += CV_SQR(t);
num -= t*templMean[k];
}
wndMean2 *= invArea;
}
if( isNormed || numType == 2 )
{
for( k = 0; k < cn; k++ )
{
t = q0[idx2+k] - q1[idx2+k] - q2[idx2+k] + q3[idx2+k];
wndSum2 += t;
}
if( numType == 2 )
num = wndSum2 - 2*num + templSum2;
}
if( isNormed )
{
t = sqrt(MAX(wndSum2 - wndMean2,0))*templNorm;
if( fabs(num) < t )
num /= t;
else if( fabs(num) < t*1.125 )
num = num > 0 ? 1 : -1;
else
num = method != CV_TM_SQDIFF_NORMED ? 0 : 1;
}
rrow[j] = (float)num;
}
}
}
