MinEigenVal output normalization depends from derivative kernels

MinEigenVal output normalization depends from derivative kernels

Hi,

while comparing the minimum eigenvalue computed for an image using ippiEigenValsVecs_8u32f_C1R / ippiEigenValsVecs_32f_C1R with the same feature evaluated from scratch with the given formulas, it seems that the output is being normalized by a factor that depends from the employed derivative kernel. In particular, results are normalized by 4080 (255 << 4) and 65280 (255 << 8) in case of Sobel 3x3 and Sobel 5x5, respectively. The normalization factor for Scharr 3x3 seems about 6661.23.... where does this values come from? Anybody can answer?

thanks

PS: i am using IPP 6.0

3 Beiträge / 0 neu
Letzter Beitrag
Nähere Informationen zur Compiler-Optimierung finden Sie in unserem Optimierungshinweis.

IPP computer vision functionality corresponds to OpenCV one – so normalization is done by the the same way as in OpenCV (in order to be compatible with OpenCV), so Normalization is as the following ways

…………
denom = 1.0f/(Ipp32f)((1<<(2*apertureSize-2))*255);
if (kernType==ippKernelSobel) {
…………………….
} else { // kernType==ippKernelScharr
denom *= 0.6125f;
………………………………
}
……………
calcValues (bufXXomp,bufYYomp,bufXYomp,step32f,pEigenVV,eigStep,roiSize,denom,sign)

Many thanks, compatibility with OpenCV makes sense.

Kommentar hinterlassen

Bitte anmelden, um einen Kommentar hinzuzufügen. Sie sind noch nicht Mitglied? Jetzt teilnehmen