Does anyone know of a fairly fast algorithm for generating a "hash" of images which behaves like soundex, in that similar images have similar values? I know that's not possible in general, but I'm particularly interested in text - so what I really want is something that says "boo" is more like "bon" than "box" (in the character set I am using to type this, at least). You can assume no background noise, good orientation etc. [more inside]
posted by andrew cooke on Jul 28, 2005 - 17 answers