On Thu, Jan 15, 2004 at 11:11:34 -0600, Bob Apthorpe wrote: > You could just normalize the images to a 64x64 bitmap with a reduced color > depth (12bpp; ~48kb uncompressed) and store that; there are probably some > optical cross-correlation techniques you can use to measure how similar > the images are. Dropping the resolution makes it more difficult to tweak > the image to generate different hashes and keeps the database size > manageable.
Last year, I worked on a project to do motion detection between 2 jpegs and extra features like drawing a box around the movement or issuing control instructions to a mobile camera. obviously most of these features aren't needed, but its the motion detection code that may be useful. it took us about 6 weeks to perfect it. I will try and dig it out if anyone wants to see it. > Though I'd really like to see optical Bayes poisoning attempts: "Check it > out - a blender, a map of Zaire, Winston Churchill, and two purple > upside-down naked chicks - it's Warhol Spam!" I don't care if it's poisoning, I want a copy of that pic :) -- ----------------------------------------- + Mat Harrison | [EMAIL PROTECTED] + | England, UK | [EMAIL PROTECTED] | |--------------+--------------------------| + http://www.genestate.com + ---------------------------------------- Yes, of course it's the right cabl [le0: NO CARRIER]
pgp00000.pgp
Description: PGP signature