Is Eigenfaces best for facial comparison? [closed]

Question

I'm working with a set of images and computer vision. I want to compare a new image (a new face) to the set of images and determine which image in the set it is closest to. I don't need it to determine which is an actual match, just something like the probability that the new image matches a particular one in the set.

So far, I've been generating Eigenfaces with the set of images I want to compare to. Is this the best method though, considering I don't want to see if there is an actual complete match?

Thanks for the help!

Answer 1

If you have good control over the lighting/pose/image registration, eigenfaces would indeed work well.

In a more unconstrained environment, I'd recommend an approach which finds relevant facial landmarks like eyes/nose/mouth corners etc. and produces descriptors around those regions. Some machine learning can then be used to distinguish between individuals.

The Oxford Visual Geometry Group has a good example of this here. This is rather old but has code included and is thus a good starting point for you to hack on. If you want something that works ASAP, then consider buying a commercial solution like PittPatt.