"Skin" detecting algorithm saves kids from Chatroulette

As with most technological advancements it is often just a matter of time before their potential for positive impact is adapted by the more depraved among us. 

Of course it is not the technology to blame, rather how us humans choose to use it. For example Microsoft’s Kinect system has been utilised for the greater good, namely assisting in remote surgery, as well as purely for cheap thrills.

But perhaps nowhere else is this point more clear than video chat randomiser website, Chatroulette. 

What could be used as an open way to instantly meet and talk with random people across the world has, as anyone who has visited can attest, become an excuse for men from around the world to proudly pull on their todgers.

Of course there are more lewd and shocking places around the internet, but the problem with Chatroulette is that the site, amounting to 6.3 million visitors in July 2010, is easily accessible by, and popular with, kids.

Xinyu Xing, from the University of Colorado at Boulder, says that there is no easy way to police Chatroulette, so it’s tough to put a block on wide-eyed innocents being exposed to frantic, furious exhibitionist masturbators, says the MIT Technology Review but not in those words.

“Our observations on a typical Saturday night indicate that as many as 20-30 percent of Chatroulette users are minors,” he says.

However Xing and his team believe they have found an answer to the ‘flasher’ problem in an algorithm that is able to spot offenders before giving them the boot.

A previous method to try to catch flashers has been to use a crowdsourcing mechanism in which users report repeat offenders whose video feed can subsequently be evaluated by trained individuals before being stopped.

However this is time consuming, and considering the high flow of traffic it is impractical.

Existing algorithms that have been desgind to detect pornographic content are also unable to work effectively as they find it difficult to distinguish between the skin on show and backgrounds in many cases, according to Xing.

In order to solve this, Xing and his team developed their own algorithm, dubbed SafeVChat, which has been tested on 20,000 still images taken from the site as supplied by founder Andrey Ternovskiy.

What is different about Xing’s approach is that it combines skin detection with motion detection that compares sequential frames to decide whether the ‘skin’ is moving, before analysing the image using several criteria to judge whether the image is acceptable or not.

The technology is also able to determine between the skin on faces and on er other parts of the body using face, eye and nose detectors, before the results are fused and are classified as normal or offensive, having been trained on the dataset.

Apparently the software has worked significantly better than a commercial pornographic image detector programme called PicBlock, and is so effective that Chatroulette has started using SafeVChat on its site.