The Science Behind Software That IDs Dudes Who Flash Dongs On Chatroulette
Posted by Justin on January 26th, 2011
You know, if we have a machine that can destroy legendary trivia heads at Jeopardy it’s high time we had a piece of software that could spot some creep trying to pull out their 4 Wood on sites like Chatroulette. But how?
A new research paper out of Cornell explains the symphony of algorithms behind identifying some random Johnson while adjusting for varying light, skin tone and image composition. For example, in a video demonstrating the program a picture of a couple lying together topless on a screen without exposing any naughty bits did not trigger as low of a rating a weirdo lifting their shirt to grip their tallywhacker for the cam. It also recognized low light and static images.
Pretty amazing stuff and a clever solution to a har… err… difficult problem.
[SafeVchat: Detecting Obscene Content and Misbehaving Users in Online Video Chat Services]
via [Improbable Research]
Twitter this
January 27th, 2011 at 5:28 am
Just wanted to say what great web-site you have.
February 17th, 2011 at 5:42 am
I love this post very much. I will certainly be back.