[Mimedefang] Image blocking idea

[Paul Murphy]

> Here's an idea for blocking image spam:  What about taking the idea of 
> SURBL and DNSRBls and extending it to images.  My proposal is to hash the 
> image and do a DNS query using the hash value and domain hosting the image 
> RBL.

You'd need to do some more complex processing, as the image can differ by one
pixel and have a different hash from the original.  Imagine an image of
300x200 where the last two lines were reserved as a unique image ID - you
have 400 bits of randomness which the average eye will fail to notice or will
filter out, and which can easily be manipulated by the image creation
software.

One possible approach is to split the image into 16 parts (a 4x4 grid), and
compute something like the average colour of each panel and then a hash/value
for the coverage of the panel when the colourmap is reduced to 2 bits.
However, there will always be a way for spammers to vary images which
confuses the computational filter but which the eye turns into readable text.
The same applies to OCR, but for the moment it has some value.

Imagine an animated GIF file where the first frame says "Click here for", the
second shows a V, the third an I, the fourth an A, then G, R and A - the
brain can join the dots, but can any software?

Paul.

-- 
No virus found in this outgoing message.
Checked by AVG Free Edition.
Version: 7.1.385 / Virus Database: 268.4.4/319 - Release Date: 19/04/2006