Monitoring every image posted on a network not only raises delicate privacy concerns, but also becomes an almost impossible challenge. This has led social networks and similar environments to rely on algorithms to recognize restricted content, such as nudity. IsItNude is a new algorithm focused on nudity detection, serving as a demonstration for its developer Algorithmia.
The Facebook Incident
In November 2012, Facebook censored an image of a woman in a bathtub (read bathtub or tub). The moderators, probably assisted by an algorithm, blocked the photo because they mistook the woman's left elbow for an exposed nipple. The image was quickly restored with an appropriate apology, but it's an excellent example of what false positives can cause in a moderation system. The computing world has a great interest in limiting exposure to adult content, including nudity. In some cases internal policies are cited, but in others, it's just a question of time and place.
How IsItNude Works
Considering false positives, is it a good idea to let an algorithm do the heavy lifting of detecting nudity? The folks at Algorithmia, responsible for the IsItNude site, believe so. IsItNude's operation is very simple: Paste the URL of the image you want to analyze, and you'll receive the result in seconds. Needless to say, your judgment on the image will be much faster and more accurate than IsItNude's, but the truly interesting thing here is how the algorithm works (face detection, nose recognition, skin tone changes, and polygon analysis), and the fact that any interested developer can integrate this function into their projects through the Algorithmia API.
Accuracy Concerns
Now, we come to "the" question: How good is IsItNude? According to my tests, let's say at the moment I wouldn't trust it with the security of any platform. I've noticed problems with partial nudity and illustrations, and it's incompatible with black-and-white images, so its developers have quite a bit of work on their hands. Of course, a bad result isn't enough to dismiss its potential. The company has received over two million dollars in investments, which represents a big bet on the growth of this technology.
