static bool matchTemplate_SQDIFF_NORMED(InputArray _image, InputArray _templ, OutputArray _result)
{
matchTemplate(_image, _templ, _result, CV_TM_CCORR);
int type = _image.type(), cn = CV_MAT_CN(type);
ocl::Kernel k("matchTemplate_SQDIFF_NORMED", ocl::imgproc::match_template_oclsrc,
format("-D SQDIFF_NORMED -D T=%s -D cn=%d", ocl::typeToStr(type), cn));
if (k.empty())
return false;
UMat image = _image.getUMat(), templ = _templ.getUMat();
_result.create(image.rows - templ.rows + 1, image.cols - templ.cols + 1, CV_32F);
UMat result = _result.getUMat();
UMat image_sums, image_sqsums;
integral(image.reshape(1), image_sums, image_sqsums, CV_32F, CV_32F);
UMat templ_sqsum;
if (!sumTemplate(_templ, templ_sqsum))
return false;
k.args(ocl::KernelArg::ReadOnlyNoSize(image_sqsums), ocl::KernelArg::ReadWrite(result),
templ.rows, templ.cols, ocl::KernelArg::PtrReadOnly(templ_sqsum));
size_t globalsize[2] = { result.cols, result.rows };
return k.run(2, globalsize, NULL, false);
}
static bool convolve_32F(InputArray _image, InputArray _templ, OutputArray _result)
{
_result.create(_image.rows() - _templ.rows() + 1, _image.cols() - _templ.cols() + 1, CV_32F);
if (_image.channels() == 1)
return(convolve_dft(_image, _templ, _result));
else
{
UMat image = _image.getUMat();
UMat templ = _templ.getUMat();
UMat result_(image.rows-templ.rows+1,(image.cols-templ.cols+1)*image.channels(), CV_32F);
bool ok = convolve_dft(image.reshape(1), templ.reshape(1), result_);
if (ok==false)
return false;
UMat result = _result.getUMat();
return (extractFirstChannel_32F(result_, _result, _image.channels()));
}
}
static bool ocl_norm( InputArray _src, int normType, InputArray _mask, double & result )
{
const ocl::Device & d = ocl::Device::getDefault();
#ifdef __ANDROID__
if (d.isNVidia())
return false;
#endif
const int cn = _src.channels();
if (cn > 4)
return false;
int type = _src.type(), depth = CV_MAT_DEPTH(type);
bool doubleSupport = d.doubleFPConfig() > 0,
haveMask = _mask.kind() != _InputArray::NONE;
if ( !(normType == NORM_INF || normType == NORM_L1 || normType == NORM_L2 || normType == NORM_L2SQR) ||
(!doubleSupport && depth == CV_64F))
return false;
UMat src = _src.getUMat();
if (normType == NORM_INF)
{
if (!ocl_minMaxIdx(_src, NULL, &result, NULL, NULL, _mask,
std::max(depth, CV_32S), depth != CV_8U && depth != CV_16U))
return false;
}
else if (normType == NORM_L1 || normType == NORM_L2 || normType == NORM_L2SQR)
{
Scalar sc;
bool unstype = depth == CV_8U || depth == CV_16U;
if ( !ocl_sum(haveMask ? src : src.reshape(1), sc, normType == NORM_L2 || normType == NORM_L2SQR ?
OCL_OP_SUM_SQR : (unstype ? OCL_OP_SUM : OCL_OP_SUM_ABS), _mask) )
return false;
double s = 0.0;
for (int i = 0; i < (haveMask ? cn : 1); ++i)
s += sc[i];
result = normType == NORM_L1 || normType == NORM_L2SQR ? s : std::sqrt(s);
}
return true;
}
bool forward_ocl(InputArrayOfArrays inps, OutputArrayOfArrays outs, OutputArrayOfArrays internals)
{
std::vector<UMat> inputs;
std::vector<UMat> outputs;
inps.getUMatVector(inputs);
outs.getUMatVector(outputs);
for (size_t i = 0; i < inputs.size(); i++)
{
UMat srcBlob = inputs[i];
void *src_handle = inputs[i].handle(ACCESS_READ);
void *dst_handle = outputs[i].handle(ACCESS_WRITE);
if (src_handle != dst_handle)
{
MatShape outShape = shape(outputs[i]);
UMat umat = srcBlob.reshape(1, (int)outShape.size(), &outShape[0]);
umat.copyTo(outputs[i]);
}
}
outs.assign(outputs);
return true;
}
