FFT image registration fails for multiples of 8

[Harlequinade]

I notice actually now that the first element in the matrix at position (0,0) for the 16x16 matrix is 0 + 0i.

Then doing division by zero leads to NaN. Putting a matrix with one value NaN into the inverse FFT method gives out an array of NaN.

I guess I'll just replace the first element in the array at position (0,0) to some non-zero value and it'll work. I don't really understand why this is the case.

[altenbach]

Why don't you simply use the 2D cross correlation?

(Here's a very rought draft. probably needs quite a few tweaks) 😉

[Harlequinade]

You're absolutely right, but the whole pursuit was to compare the two with each other. See which one is faster or better for my set of data and if I can do things better than NI. 

Another objective was to put the algorithm on the FPGA.