> > On Aug 2, 2006, at 3:03 AM, Matthias Keller wrote: > >>>> will it not be much faster just to make a md5 sum on the image file >>>> without >>>> thinking if it a appel or orange ? :-) >>> Yes, but just taking a straight sum will be sensitive to all of those >>> small pixels which are changed by the spammers so that they have >>> different sums, but the differences aren't visible to the human eye. >>> >>> That's my point. If you drop out the lower bits of the colors, then >>> you mostly retain what is perceptible (in color ranges) to the human, >>> while losing those parts that a) the human wouldn't have noticed >>> anyway, and b) throw off your sum of the image for comparison to >>> known spam images. >> Hi >> >> You're idea is kinda interesting, but what would you do about a pic >> with white background, black font and some random black noise on it ? > > > Yeah, my strategy fights hidden pixel variations, but not overt ones. > making the image actually appear grainy/noisy to the human eye, with > different grain/noise for each spam, still gets past my strategy. > > Maybe I'm not getting the obvious, but what about using something like Perl::Magick to convert a given image into B/W? I mean, ImageMagick is made for things like that... Shrinking it to, say, a quarter of it's original size would take care of at least many random noise pixels.
Dirk
