Thursday, December 08, 2005

Are images SPAM's latest avatar?

I'm seeing more spams using images. Basically the spammer embeds the text in a .gif file. So most of the spam filtering software cannot make out from the text portion of the mail, if it is spam or not. See the spam picture that I got today.

I think the spam software these days use some form of AI -- machine learning using known patterns of good text and bad ones. So when an unknown mail arrives, it matches the text to the nearest known "point" in the "good mail space" and in the "bad mail space". If it is too close to the bad mail, it flags it as spam.

But can software handle spam inside images? We know computers aren't (yet) good at image recognition. In fact Amazon's HITs and mturk are built over this fact. Considering machines can easily generate images which they can't understand but humans can, is this the next step in bypassing spam filters?

Wondering how yahoo/gmail are weeding out such stuff... btw google image search on 'image spam' turns out lot of eatable spam pics! ;)


Post a Comment

<< Home