Mon Oct 14 19:59:41 UTC 2002
I'm starting to see more and more of this... spam which consists of simply an image attachment. While this costs more bandwidth for the spammer to send (that is, if they are using their bandwidth and not relay raping), this provides a means to defeat most filters as images can be perturbed very easily and in many different ways. Slightly blur, sharpen, color change, or move elements in an image and it can be perturbed into an endless array of possible binary representations. It seems to me that this breaks all the fuzzy checksums that DCC is capable of. Is there any plan to integrate a fuzzy image checksum in the future? It seems like this would not be *that* hard, as GPL/BSD code for fuzzy image fingerprinting has probably already been written somewhere. It would just have to be plugged in. I know google and some web filters use this kind of technology to categorize and group images by similarity. Of course, you could automatically bulk-mark all mail that consists only of an image or a small quantity of HTML with images. Such things are normally not sent as part of normal communication.
More information about the DCC