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

added support of CV_32F & CV_TM_SQDIFF into gpu::matchTemplate

Browse Source
This commit is contained in:
Alexey Spizhevoy
2010-12-08 13:12:12 +00:00
parent 40304721a7
commit b1c5b9293e
3 changed files with 112 additions and 43 deletions
Download Patch File Download Diff File Expand all files Collapse all files

49
modules/gpu/src/cuda/match_template.cu
View File

@@ -50,6 +50,7 @@ using namespace cv::gpu;
namespace cv { namespace gpu { namespace imgproc {
texture<unsigned char, 2> imageTex_8U;
texture<unsigned char, 2> templTex_8U;
@@ -98,6 +99,54 @@ void matchTemplate_8U_SQDIFF(const DevMem2D image, const DevMem2D templ, DevMem2
}
texture<float, 2> imageTex_32F;
texture<float, 2> templTex_32F;
__global__ void matchTemplateKernel_32F_SQDIFF(int w, int h, DevMem2Df result)
{
int x = blockDim.x * blockIdx.x + threadIdx.x;
int y = blockDim.y * blockIdx.y + threadIdx.y;
if (x < result.cols && y < result.rows)
{
float sum = 0.f;
float delta;
for (int i = 0; i < h; ++i)
{
for (int j = 0; j < w; ++j)
{
delta = tex2D(imageTex_32F, x + j, y + i) -
tex2D(templTex_32F, j, i);
sum += delta * delta;
}
}
result.ptr(y)[x] = sum;
}
}
void matchTemplate_32F_SQDIFF(const DevMem2D image, const DevMem2D templ, DevMem2Df result)
{
dim3 threads(32, 8);
dim3 grid(divUp(image.cols - templ.cols + 1, threads.x),
divUp(image.rows - templ.rows + 1, threads.y));
cudaChannelFormatDesc desc = cudaCreateChannelDesc<float>();
cudaBindTexture2D(0, imageTex_32F, image.data, desc, image.cols, image.rows, image.step);
cudaBindTexture2D(0, templTex_32F, templ.data, desc, templ.cols, templ.rows, templ.step);
imageTex_8U.filterMode = cudaFilterModePoint;
templTex_8U.filterMode = cudaFilterModePoint;
matchTemplateKernel_32F_SQDIFF<<<grid, threads>>>(templ.cols, templ.rows, result);
cudaSafeCall(cudaThreadSynchronize());
cudaSafeCall(cudaUnbindTexture(imageTex_32F));
cudaSafeCall(cudaUnbindTexture(templTex_32F));
}
__global__ void multiplyAndNormalizeSpectsKernel(int n, float scale, const cufftComplex* a,
const cufftComplex* b, cufftComplex* c)
{

12
modules/gpu/src/match_template.cpp
View File

@@ -61,6 +61,7 @@ namespace cv { namespace gpu { namespace imgproc
void multiplyAndNormalizeSpects(int n, float scale, const cufftComplex* a,
const cufftComplex* b, cufftComplex* c);
void matchTemplate_8U_SQDIFF(const DevMem2D image, const DevMem2D templ, DevMem2Df result);
void matchTemplate_32F_SQDIFF(const DevMem2D image, const DevMem2D templ, DevMem2Df result);
}}}
@@ -92,6 +93,14 @@ namespace
imgproc::matchTemplate_8U_SQDIFF(image, templ, result);
}
template <>
void matchTemplate<CV_32F, CV_TM_SQDIFF>(const GpuMat& image, const GpuMat& templ, GpuMat& result)
{
result.create(image.rows - templ.rows + 1, image.cols - templ.cols + 1, CV_32F);
imgproc::matchTemplate_32F_SQDIFF(image, templ, result);
}
#ifdef BLOCK_VERSION
template <>
@@ -243,7 +252,8 @@ void cv::gpu::matchTemplate(const GpuMat& image, const GpuMat& templ, GpuMat& re
typedef void (*Caller)(const GpuMat&, const GpuMat&, GpuMat&);
static const Caller callers8U[] = { ::matchTemplate<CV_8U, CV_TM_SQDIFF>, 0, 0, 0, 0, 0 };
static const Caller callers32F[] = { 0, 0, ::matchTemplate<CV_32F, CV_TM_CCORR>, 0, 0, 0 };
static const Caller callers32F[] = { ::matchTemplate<CV_32F, CV_TM_SQDIFF>, 0,
::matchTemplate<CV_32F, CV_TM_CCORR>, 0, 0, 0 };
const Caller* callers;
switch (image.type())