ippiCrossCorrSame_NormLevel with non square templates

[seddona]

Hi,

Is it possible to use ippiCrossCorrSame_NormLevel with non square templates(parallelogram etc)? I can't seem to find a masked version.

Thanks,

[Chao_Y_Intel]

nomasked version so far. Can the non square templates be extended into a square one by adding 0 element ?


Thanks,
Chao

[seddona]

yes this is what I have done, but marking the unused area with 0's or 255's will effect the final correlation result I belive. Possibly there is an out of bounds number I could use on one of the higher bit depth functions?

[seddona]

Quoting - Chao Yu (Intel)

nomasked version so far. Can the non square templates be extended into a square one by adding 0 element ?


Thanks,
Chao

So does the ippiCrossCorrSame function use FFT based correlation? Perhaps above a certain template size?

[Vladimir_Dudnik]

There is comment from our expert:

Yes, it is FFT based for all template sizes (as FFT is more efficient for templates > ~10x10 than direct algorithm and it seems that such small templates are out of interest; CrossCorrValid version has direct implementation for rather small templates and switches to FFT based algorithm at some empirical criterion). Padding of complex template to rectangle with zeroes (or any other constant) doesn't affect resulting matrix of correlation coefficients (especially normalized-level version) as maximum value will be any way at the point of maximum match.

Regards,
Vladimir

[seddona]

thanks for the info.