I've been watching the whole bruhaha at Seattle911.com with some interest. This is a homegrown website that takes live 911 feeds and puts them on a Google mashup. Cute and clever use of technology. The Seattle Fire Department responded by changing the logs to image format rather than text.
That's the background. Reading some of the articles pointed me to 'gocr' which is a free OCR package. Now this is useful - I wasn't earlier aware of its existence. It basically takes an image and tries to distinguish text in the image and gives you the text. If you saw my article on comment spam, you'll realize that 'captcha' images to prevent spam are doomed. This is where you type into a box the letters you see in the picture. Most of these are annoying anyway, but it's pretty hard to get them through a clever command-line driven OCR program. If you make it so hard to read that gocr can't read it, chances are that none of your audience will be able to either.
But I have an even deeper interest in this stuff. Gocr is a framework for finding recognizable stuff in images. Something the world has needed for a while now is something that can filter porn. In theory there isn't much difference between distinguishing the letter 'b' in a picture (in any of 600 different fonts) and say a breast (in any of 600 different sizes/shapes). I'm being polite. Any anatomical feature.
Some folks worked on this problem back in the '80's, correlating the prevalance of what could be termed 'skin tones' in an image.
The tools and concepts are out there. It shouldn't take much more than a man month or three to put them together into a porn filter. There's probably a market for such a thing.
OK, gocr is probably encumbered with the GNU General Public License. So maybe there isn't much of a market unless one just uses the general pattern recognition concepts (but not the code) and starts from scratch. I don't have anything against the GPL. It serves its purpose, but it does make it hard to re-use code in the workplace. It's a bummer to always have to start from scratch, when the software already exists and has been pretty much debugged. If I'm releasing code into the public domain, I always use either the Berkeley/Stanford license or no license at all. Free, no warranty, blah blah. The GPL is basically a self-replicating virus - which was written by lawyers instead of geeks.