ippiCrossCorrValid_NormLevel_8u_C1RSfs

[shyweissman]

I am trying to use cross correlation routine and I get zeros at part of the result matrix. also on opencv code does not. It seems that all parameters are ok.

Ipp8u* imgMatr, max_val;

imgMatr = ippiMalloc_8u_C1(64,64,&stepBytes );

p_img_src = (Ipp8u*)imgCurr8->imageData+rectSrc.x + start_x +

(rectSrc.y + start_y)*imgCurr8->width;

p_img_templ = (Ipp8u*)imgMatch8->imageData+rectMatch.x + rectMatch.y*imgMatch8->width;

IppStatus Corr_status = ippiCrossCorrValid_NormLevel_8u_C1RSfs(p_img_src,imgCurr8->width, srcRoiSize,

p_img_templ, imgMatch8->width,

tplRoiSize, (Ipp8u*)imgMatr, stepBytes, -8);

ippiMaxIndx_8u_C1R(imgMatr, stepBytes, SizeMat, &max_val, max_x, max_y);

[Vladimir_Dudnik]

Hi,

there is answer from our expert:

Yes, of course, this code leads to zero matrix. As it is mentioned in ippiman, all correlation coefficients are obtained in 32f data type and 8u output means scaling of 32f coefficients to 8u data type range with appropriate scale factor. For the full 8u range (0-255)you should use scaleFactor = -8.

Regards,
Vladimir

[Intel_C_Intel]

Hi,

Another note - IPP cross correlation is calculated after mean subtraction (see manual), so you'll get zeroes for constant images

Alexander