Phase correlation without normalization




@axtimwalde @StephanPreibisch

I am using the imglib phasecorrelation algorithm. For me, I find that such cross-correlation based algorithms often “jump” to bright regions (not sure I am explaining myself here properly…). Anyway, I was wondering whether a similarity measure without normalization, e.g., simply the mean-square difference (without dividing each image by its standard deviation) would be more robust. And basically I was wondering whether the phasecorrelation algorithm provides this option at all.



Hi @Christian_Tischer,

I guess you mean the mpicbg.imglib.algorithm.fft.PhaseCorrelation class (which is part of legacy-imglib1). Can you specify how you’re using it on your images? Also, looking at the source code might help you to find out how it is doing normalization and maybe allow you to cherry-pick just the code that calculates correlation.

Maybe the ImgLib2 maintainers can comment on whether there is a good (and possibly improved) replacement for this class available in ImgLib2. (I added the appropriate tags to your post.)