Hacker News Reader

Comment by michael_dorfman | Hacker News Reader

Setting aside the business model issues for a moment: I'm really intrigued by the idea of penis-recognition software. Now, that's a branch of AI I never even considered before, and already my mind is racing through the algorithmic approaches...

View on HN

If they get the penis-recognition AI working, they can split the site in two, chattroulette and penisroulette, and automatically route videos to the correct site. I'm honestly not sure which site would become the more popular though.

I think the penis showing men would just disguise their penises. So we would see penises with little wigs and glasses on. I would like to see the penis AI cope with that!

Would you really?

"The Berkeley-Iowa Naked People Finder" describes one approach, which seems to first classify pixels as skin/not skin by color, then identifying "limbs" from the pixel regions.

http://www.cs.hmc.edu/~fleck/naked.html

Some other people have worked on better skin classifiers:

http://www.computer.org/portal/web/csdl/doi/10.1109/CVPR.199...

http://portal.acm.org/citation.cfm?id=1240564

Looks like a hot topic. :)

I know some people who worked on a 'porn detector' project at a University. Their main job was to tag the training and test sets by labelling breasts and other naughty bits. Fun job! ;)

They will find away around it. Thousands of pervs translate into hundred of thousands of man hours to find away around the filter. A handful of developers/mathematicians can't keep up with that.

Web filtering software and even Google Images employ techniques that find porn on images and videos. While I'm not sure about its details, it has been done for a long time and I think it's fairly accurate (even if its false positives are a issue in some cases).

Reply on news.ycombinator.com