Spam in images

[Tom Anderson]

John G Walker wrote:
> Fair point. That's why I insisted that one word should be enough for
> the email message to go through the normal (statistical) routines.
> 
> But the problem is that you can't apply statistical techniques to a
> population of zero size. At the moment, the way things work, a message
> of zero words gets automatically classed as ham. Hence the discussion
> about analysing images.
> 
> What I'm suggesting is that whether a zero-word message is classified
> as ham or spam should be under the control of the user, via an option
> on the command line. The default, to enable backwards compatibility,
> should be ham,

That's simply not true.  All emails have dozens of tokens in the header 
on which to filter.  Having no body text is indicative of nothing.

Tom