how to compare two image which should neglect the lighting changes

i want to compare two images .... i previously used subtraction to compare images but the problem now is even with a slight lighting variation the program is showing the change has been detected .. so plz suggest me how to neglect the this lighting problem for comparing my image....

Answer 1

[Image Analyst]

Well subtraction did detect a difference didn't it? Apparently though, it wasn't a difference you were interested in. But I don't really know how you want to compare them. So maybe you should look at SSIM, CBIR, Homomorphic filtering, or SURF ( http://en.wikipedia.org/wiki/SURF) They're all in wikipedia. Or just browse http://iris.usc.edu/Vision-Notes/bibliography/contents.html til you find something that looks like what you want to do.

Answer 2

[Geoff]

Try using a threshold. Set a minimum amount of change to denote movement. Anything less than that value is noise. Lip movement presumably gives big deltas, whereas lighting changes from one frame to the next ought to have smaller deltas.

You could also try using more relevant colour encodings (if you are currently working in RGB), such as YUV. The goal here being that you want to separate changes in lighting from changes in, say, chroma (although in your case, the chroma of lips versus skin is might not be enough to offset apparent chroma changes due to lighting).

Another trick is to do the difference across older frames. But I dunno. Difference images don't work very well if your target isn't moving very fast.

You might do better doing feature tracking.