In the context of handwriting recognition, what are eigenimages? Is there a decent explanation of them anywhere? I searched Wikipedia, which is often a good place to start for this sort of thing, and nothing came up. Google doesn’t help much, either.
Thanks
This is just a wild guess (i.e. I don’t really know what I’m talking about), but I’d say an eigenimage is an image for a handwritten letter approximated by treating an input “example” set of handwritten letters as an eigenspace. I suppose this would allow one to sort a predefined dictionary of patterns based on some predefined criteria of similarity to an input image.
Another wild guess here.
But since eigenvalues (and their corresponding eigenvectors) describe preferred directions of rotation of physical bodies, perhaps eigenimages are some sort of preferred patterns (or images) of a person’s handwriting habits?
“Eigen” means “self” in German, approximately. Make of that what you will.
Best guess: they’re treating the collection of all images as a vector space and looking for the eigenvectors (since vectors here are images, “eigenimages”) of some operator on that space. Physicists do the same thing when they call eigenvectors in quantum mechanical contexts “eigenstates”. For eigenvectors, that you should be able to find on Wikipedia or mathworld.
Ahh, cool. Thanks. I don’t think I’ll be needing to use them, then.
I would assume that it’s very very similar to the use of eigenfaces in facial recognition software.