xcorr returning a triangular shaped plot

I'm currently trying to find a the time delay between peaks in CO2 concentration and the peaks in intensity gained from and MRI scan. But so far when I use xcorr and plot the results I just get a triangular shaped peak at 0 lag which I know can't be right.

Has anyone any idea what I am doing wrong?

So far my code it pretty simple:

 a=load(CO2 file)
 b=load(MRI file)
 T=a/norm(a)
 Y=b/norm(b)
 time=92 % There are 92 data point
 figure 1
 plot(time, T, time, Y)% Just to see what the normalised data looks like on the same graph

 C=xcorr(a, b, 20, 'coeff') % Only interested in 20 lags either way
 time2=-20:20
 plot(time2, C) % Results in a triangular-shaped plot

http://tinypic.com/r/2v9utdk/6

[ChristianW]

Can you normalise a and b with T=(a-mean(a))/std(a) and show the plot? Alternatively give us some data.

[Marek]

Ahh no need, normalising the data in the way you described has solved the problem now and I'm getting more reasonable results. Thanks!!

[TJ]

Could you elaborate on what fixed your problem? How did you normalize your data and what did that do?

Answer 1

[Image Analyst]

Maybe you should use normalized cross correlation, done by normxcorr2() in the Image Processing Toolbox (IPT). IPT functions also work with 1D signals as well as 2D images and sometimes 3 dimensions or more.

Often non-periodic signals will correlate and give a roughly triangular, or mountain-shaped, result so what you're seeing is not uncommon. In fact the correlation of two uniform signals will give a perfect triangle - just consider how the area of overlap changes as two boxes slide past each other.