To reduced influence of amplitude shifts and centre brightspot, we need to normalise the cross correlation coefficient. This is tricky because each element in the correlogram requires a different denominator and can't be done in FFT (easily) -> there are ways to do this fast however with a fast precalc (eg https://www.researchgate.net/publication/2378357_Fast_Normalized_Cross-Correlation).
There is another problem with the upsampled DFT - we need to get the normalisation denominators in the upsampled region as opposed to whole correlogram.
To reduced influence of amplitude shifts and centre brightspot, we need to normalise the cross correlation coefficient. This is tricky because each element in the correlogram requires a different denominator and can't be done in FFT (easily) -> there are ways to do this fast however with a fast precalc (eg https://www.researchgate.net/publication/2378357_Fast_Normalized_Cross-Correlation).
There is another problem with the upsampled DFT - we need to get the normalisation denominators in the upsampled region as opposed to whole correlogram.