Bilateral Filter

[gkguru]

Hi,

I am having a little trouble understanding and initializing bilateral filter.

First of all, can anyone explain what the argument 'stepInKernel' in the method FilterBilateralInit mean?
Also, how does one initialize the IppiFilterBilateralSpec structure/allocate external buffer? The information provided in the manual is a bit confusing.

I get a run time error when running this piece of code.

status = ippiFilterBilateralGetBufSize_8u_C1R(ippiFilterBilateralGauss, imgSize, kernelSize, &buffer);
pSpec = (IppiFilterBilateralSpec*)malloc(buffer);
status = ippiFilterBilateralInit_8u_C1R(ippiFilterBilateralGauss, kernelSize, 1.0, 1.0, stepInKernel, pSpec);
status = ippiFilterBilateral_8u_C1R(srcGrayImg, srcGrayStep, dstGrayImg,dstGrayStep,imgSize, kernelSize, pSpec);

Thanks
Guru

[gkguru]

By bad...Forgot to add borders. So the filter is now working.

But I still dont know what stepInKernel - "Processing" step in the filter kernel means?


Thanks
Guru

[malikcoates]

I'd also like an answer to this. The stepInKernel parameter seem to be undocumented.

[Ivan_Z_Intel]

The filter processes only the pixels what have the kernel coordinates multiple of stepInKernel. The coordinates of center are (0,0).

So if stepInKernel is equal 1 then all pixels take part in processing.

For example kernel is 5x5.

If stepInKernel == 1 then

the pixels in processing:

(-2,-2) ,(-1,-2), (0,-2),(1,-2),(2,-2),

(-2,-1) ,(-1,-1), (0,-1),(1,-1),(2,-1),

(-2,0) ,(-1,0), (0,0),(1,0),(2,0),

(-2,1) ,(-1,1), (0,1),(1,1),(2,1),

(-2,2) ,(-1,2), (0,2),(1,2),(2,2).

If stepInKernel == 2 then

(-2,-2) , (0,-2),(2,-2),

(-2,0) , (0,0),(2,0),

(-2,2) , (0,2),(2,2).

PS

Such be better:

status = ippiFilterBilateralGetBufSize_8u_C1R(ippiFilterBilateralGauss, imgSize, kernelSize, &buffer);
pSpec = (IppiFilterBilateralSpec*)ippsMalloc_8u(buffer);
status = ippiFilterBilateralInit_8u_C1R(ippiFilterBilateralGauss, kernelSize, 1.0, 1.0, stepInKernel, pSpec);
status = ippiFilterBilateral_8u_C1R(srcGrayImg, srcGrayStep, dstGrayImg,dstGrayStep,imgSize, kernelSize, pSpec);

[malikcoates]

Thank you that is very helpful.

One more question: What is the relationship between kernelSize, and the posSquareSigma?

In the above example we find:
status = ippiFilterBilateralInit_8u_C1R(ippiFilterBilateralGauss, kernelSize, 1.0, 1.0, stepInKernel, pSpec);

I'm thinking that the kernelSize should typically be sqrt(posSquareSigma)*stepInKernel*2+1 and that higher values would have no effect and lower values would be like using a lower posSquareSigma? So, if posSquareSigma and stepInKernel are both 1, then I assume we would want a kernel size of 3x3.

Or is there more going on here?

[nonamme]

please show me an example of using this function and ranges of parameters (kernelstep, kernelsize, valsquaresigma, possquaresigma). function returns noerror but the result is an exact copy of input data.
I use functions like this:

IppiSize ippiImgSize;
ippiImgSize.height = pFrameIn->GetHeight();
ippiImgSize.width = pFrameIn->GetWidth();

IppiSize ippiKernelSize;
ippiKernelSize.height = nKernelSize;
ippiKernelSize.width = nKernelSize;

int nBufferSize = 0;

IppiFilterBilateralType ippiFilterType = ippiFilterBilateralGauss;
IppStatus ippiStatus;

ippiStatus = ippiFilterBilateralGetBufSize_8u_C1R(ippiFilterType, ippiImgSize, ippiKernelSize, &nBufferSize);

IppiFilterBilateralSpec *pSpec = (IppiFilterBilateralSpec*) ippsMalloc_8u(nBufferSize);

int nStepInKernel = nKernelStep;

ippiStatus = ippiFilterBilateralInit_8u_C1R(ippiFilterType, ippiKernelSize,
fValSquareSigma, fPosSquareSigma, nStepInKernel, pSpec);

Ipp8u *pSrc = (Ipp8u*) FrameGray8SRC.GetPointer();

Ipp8u *pDst = (Ipp8u*) FrameGray8DST.GetPointer();

ippiStatus = ippiFilterBilateral_8u_C1R(pSrc, FrameGray8SRC.GetByteSpan(),
pDst, FrameGray8DST.GetByteSpan(),
ippiImgSize, ippiKernelSize, pSpec);

=====================
Oh, I find the cause of trouble: the Sigma values should be at range 1500 and more ( for 5x5 kernel) . And we should add to out image borders with size equal to kernelsize

[Vladimir_Dudnik]

Great, does that solve all your questions?

Vladimir

[Intel_C_Intel]

Is there a possibility to get a fully working example of the bilateral filter ?

What is the difference between the ippiFilterBilateralGauss and the ippiFilterBilateralGaussFast ?

What is the buffer size we should get when using image of size (M,N) with filter kernel (k,j) ?

I copied the last example in the thread and get no error - yet the image is invalid as if nothing has happened.

[Vladimir_Dudnik]

Hello,

there is example how to call IPP bilateral filter

#include 
#include 
#include "ippcore.h"
#include "ipps.h"
#include "ippi.h"
#define D_WIDTH 8
#define D_HEIGHT 8
#define D_MAX_W_KNL 5
#define D_MAX_H_?NL 5
int main( void )
{
Ipp8u *src, *src_start;
Ipp8u *dst1, *dst2, *dst3;
Ipp8u *ptmp;
int width_src, height_src;
int width_dst, height_dst;
int step_src;
int step_dst1, step_dst2, step_dst3;
int bufsize;
int stepInKernel;
IppiSize roi;
IppiSize kernel;
int i, j;
Ipp32f valsigma;
Ipp32f possigma;
IppiFilterBilateralSpec *pSpec;
width_src = D_WIDTH + D_MAX_W_KNL - 1;
height_src = D_HEIGHT + D_MAX_H_?NL - 1;
src = ippiMalloc_8u_C1( width_src, height_src, &step_src );
for ( i = 0, ptmp = src; i < height_src; i++ ) {
for ( j = 0; j < width_src; j++ ) {
ptmp = (Ipp8u)rand();
}
ptmp = (Ipp8u *)( (char *)ptmp + step_src );
}
src_start = (Ipp8u *)( (char *)src + (D_MAX_H_?NL>>1) * step_src );
src_start += (D_MAX_W_KNL>>1);
width_dst = D_WIDTH;
height_dst = D_HEIGHT;
dst1 = ippiMalloc_8u_C1( width_dst, height_dst, &step_dst1 );
dst2 = ippiMalloc_8u_C1( width_dst, height_dst, &step_dst2 );
dst3 = ippiMalloc_8u_C1( width_dst, height_dst, &step_dst3 );
roi.width = D_WIDTH;
roi.height = D_HEIGHT;
kernel.width = D_MAX_W_KNL;
kernel.height = D_MAX_H_?NL;
ippiFilterBilateralGetBufSize_8u_C1R( ippiFilterBilateralGauss, roi, kernel, &bufsize);
pSpec = (IppiFilterBilateralSpec *)ippsMalloc_8u( bufsize );
valsigma = 16.f;
possigma = 16.f;
stepInKernel = 1;
ippiFilterBilateralInit_8u_C1R( ippiFilterBilateralGauss, kernel, valsigma, possigma,
stepInKernel, pSpec );
ippiFilterBilateral_8u_C1R( src_start, step_src, dst1, step_dst1, roi, kernel, pSpec );
valsigma = 1.f;
possigma = 1.f;
stepInKernel = 1;
ippiFilterBilateralInit_8u_C1R( ippiFilterBilateralGauss, kernel, valsigma, possigma,
stepInKernel, pSpec );
ippiFilterBilateral_8u_C1R( src_start, step_src, dst2, step_dst2, roi, kernel, pSpec );
valsigma = 16.f;
possigma = 16.f;
stepInKernel = 2;
ippiFilterBilateralInit_8u_C1R( ippiFilterBilateralGauss, kernel, valsigma, possigma,
stepInKernel, pSpec );
ippiFilterBilateral_8u_C1R( src_start, step_src, dst3, step_dst3, roi, kernel, pSpec );
ippsFree( pSpec );
printf("
 Src =");
for ( i = 0, ptmp = src; i < height_src; i++ ) {
printf("
");
for ( j = 0; j < width_src; j++ ) {
printf("%3d ",ptmp);
}
ptmp = (Ipp8u *)( (char *)ptmp + step_src );
}
printf("
 Dst1 =");
for ( i = 0, ptmp = dst1; i < height_dst; i++ ) {
printf("
");
for ( j = 0; j < width_dst; j++ ) {
printf("%3d ",ptmp);
}
ptmp = (Ipp8u *)( (char *)ptmp + step_dst1 );
}
printf("
 Dst2 =");
for ( i = 0, ptmp = dst2; i < height_dst; i++ ) {
printf("
");
for ( j = 0; j < width_dst; j++ ) {
printf("%3d ",ptmp);
}
ptmp = (Ipp8u *)( 
(char *)ptmp + step_dst2 );
}
printf("
 Dst3 =");
for ( i = 0, ptmp = dst3; i < height_dst; i++ ) {
printf("
");
for ( j = 0; j < width_dst; j++ ) {
printf("%3d ",ptmp);
}
ptmp = (Ipp8u *)( (char *)ptmp + step_dst3 );
}
ippiFree(dst3); 
ippiFree(dst2); 
ippiFree(dst1); 
ippsFree(src); 
return 0;
}

Regards,
Vladimir

[long__bingquan]

Community Admin wrote:

Is there a possibility to get a fully working example of the bilateral filter ?

What is the difference between the ippiFilterBilateralGauss and the ippiFilterBilateralGaussFast ?

What is the buffer size we should get when using image of size (M,N) with filter kernel (k,j) ?

I copied the last example in the thread and get no error - yet the image is invalid as if nothing has happened.

Have u fixed the problem u described in the last line?

[long__bingquan]

Community Admin wrote:

Is there a possibility to get a fully working example of the bilateral filter ?

What is the difference between the ippiFilterBilateralGauss and the ippiFilterBilateralGaussFast ?

What is the buffer size we should get when using image of size (M,N) with filter kernel (k,j) ?

I copied the last example in the thread and get no error - yet the image is invalid as if nothing has happened.

Have you fixed the problem ? I meet the same.