topic Re: weighted cross correlation in Intel® Integrated Performance Primitives

weighted cross correlation

j_jack_wang — Wed, 21 Jun 2006 01:47:41 GMT

Is there a function that does weighted cross correlation or weighted correlation coefficient? Anybody know if there is a software package other than ipp that does it?

TIA

jack

Re: weighted cross correlation

Vladimir_Dudnik — Tue, 04 Jul 2006 01:20:06 GMT

Hello,

there is answer from our expert

We provide normalized cross correlation function and it iscaller'sresponsibility to provide weighted signal as input for this function there are so many ways of weighting signal and differentiation it from noise, that they cant be combined in one primitive function.

Regards,
Vladimir

Re: weighted cross correlation

j_jack_wang — Wed, 13 Sep 2006 04:55:16 GMT

Hi Vladimir,

Thanks for replying to my earlier question.

What I want to do first is to have some "don't care pixels" in template matching using Normalized Correlation Coefficient. One step further than that is weighing certain pixels more than others. Merely filtering the template and picture before calling the IPP function doesn't seem to give enough freedom to achieve what I want.

Jack

Re: weighted cross correlation

j_jack_wang — Wed, 13 Sep 2006 04:56:04 GMT

Re: weighted cross correlation

Vladimir_Dudnik — Sat, 16 Sep 2006 02:13:19 GMT

Hello Jack

currently we do not support such functionality, but it is possible that in future we can add cross-correlation function with mask, so you should be able to create mask for your "don't care" pixels. It might be useful for not rectangular templates.

Regards,
Vladimir

Re: weighted cross correlation

j_jack_wang — Thu, 21 Sep 2006 05:08:34 GMT

Thanks Vladimir,

Please keep me posted if this functionality is to be added. It willdefinitely be useful for non-rectangular templates. "Don't care pixels" for cross-correlation can be achieved by setting them to zero (filtering before calling template matching) or average intensity level (for correlation coefficient) in the template. However, weighted correlation coefficient needs more than a mask as input.

Jack

Re: weighted cross correlation

Vladimir_Dudnik — Thu, 12 Oct 2006 16:48:55 GMT

Hi Jack,

sure and also please take attention to What's new section for the further IPP releases

Regards,
Vladimir