Forums

cross correlation gives wrong peak location

Started by vondykes February 4, 2008
Hi guys,

I have a problem of doing cross correlation in matlab. What I'm trying to
do is calculate the similarity of image in order to get a location of mis
matches. I do it by 'xcorr' function and compare with FFT correlation so
as to compare the result. What I have done is about to be just fine, but I
just realised that when I took a very short signal to be correlated with
another signal, then the peak may not always be detected at where it
should be. I've found out that it was because if the correlation resulting
in small height of peak then there might be other location where the data
has the peak higher than the peak occurs from cross correlation.

To make is clear, this is the code I've tried to do;

a = 2*(rand(512,1)-0.5);
b = rand(512,1);

for i = 1:1024;
    b(i) = 0;
end;

i = 1;
for h = 101:103;
    b(i) = a(h);
    i = i+1;
end;

c=rand(512,1);
for g=1:1024;
    c(g)=0;
end;

%xcorr

c=xcorr(a,b);

%FFT

L=max(length(a)-1,length(b)-1);
NL=2*L;
x=fft(a,NL);
y=fft(b,NL);

z=fftshift(real(ifft(x.*conj(y))));


As you can see, when the signal copied from a into the origin of b is very
small, it can be a problem of peak detection, and I don't know how to solve
this because I think I have set the length correctly, and the result of the
correlation is exactly the same for both xcorr and FFT correlation,
including the ones with peak detection problem. Please help me figure this
out, Thanks.

regards,
Dykes