I would like to add phase correlation as a metric because it would be useful in the presence of intensity mismatches. http://en.wikipedia.org/wiki/Phase_correlation The phase correlation result consists of a correlation coefficient and a (x,y) offset. I am thinking the error measure should be proportional to the correlation coefficient but inversely proportional to the distance sqrt(x^2 + y^2). So error = (corr coeff) / sqrt(x^2 + y^2) ?
Activity
nfaggian commentedon Aug 3, 2011
We should have a few more metrics in general - not very sure how we could easily do this but I would also like to add:
cross-correlation
mutual information
Need to sort out the maths.