the effective size of the correlation window

[lucia.and...@gmail.com]

Hi all, 

I use ASP's stereo command to correlate images to detect a tectonic displacement. The stereo algorithm is SGM with the final MGM, sub-pixel refinement is Bayes EM, the size of the correlation kernel is 7 x 7  pixels and the sub-pixel kernel size is 21 x 21 pixels. 

So I was wondering if the Bayes EM sub-pixel refinement implemented in ASP uses Cauchy or Gaussian weighting, and if it is Cauchy, is it correct to assume that the effective size of the correlation window would be approximately half of the sub-pixel kernel, so about 10 pixels? 

Many thanks for your help. 

Lucia 

[Oleg Alexandrov]

To the best of my knowledge and as I seem to see in the code, Bayes EM (--subpixel-mode 2) uses Gaussian weights. The Gaussian sigma is determined by the kernel width. It would take some digging and figuring out to answer your question as to what the effective size of the correlation window is. 

That method is described here: https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=5413749. 

I must say that was implemented a long time ago by other people and the details elude me. If you do need more details and some experimentation does not help figure things out, I can try looking at the code or point you to the code which has the formulas.

--
You received this message because you are subscribed to the Google Groups "Ames Stereo Pipeline Support" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ames-stereo-pipeline...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/ames-stereo-pipeline-support/576bda49-904f-4955-9eb5-2adc60495d0cn%40googlegroups.com.