A recent trend in spamming observed in Larry Seltzer's blog and elsewhere is to use very small font size characters to create an ascii art to provide spam messages.

What is interesting is that unlike the example above spammers can use legitimate email messages chosen randomly from computers they have hacked to create the ascii art. Currently filters including naive bayesian are unable to filter messages which appears as normal emails.

The only way I can think of to filter such messages is to look for effective size of the fonts in the message. And if the size is not normally humanly readable (anything smaller than 7px should be a good first guess) should be treated as spam.

Any other ideas?