Forum Discussion
Altera_Forum
Honored Contributor
16 years agoYeah, is my error from a previous example.
Normalization divisor is p=H2(0,0) where H2(u,v)=G(u,v)*conj(H(u,vv)). h2(x,y)=ifft(H2(u,v))/p. You must select an adequate word and fractional bits size because h2 values could reach height*widht*max(g(x,y))*max(h(x,y)). assuming your images pixels take values ranging from 0 to 1, you must increase integer part of fixed point numbers to log2(height*widht) and consider my previous explanation for division and square root. I consider you could ommit the square root step, the peak search can be achieved without it.