Facebook Inc (NASDAQ:FB) is working on technology that could identify human faces. The social network claims that it developed technology that could verify whether or not two faces in side-by-side photos are the same. The project called DeepFace appears to be a brave attempt to replicate human capabilities.
Accuracy similar to humans
According to Facebook Inc (NASDAQ:FB), the technology is 97.25% accurate, which is only marginally below the score of 97.5% scored by humans in the same standardized test. The technology from Facebook maps out 3D facial features; after the mapping, it creates a flat model and filters it by color to characterize specific facial elements. To make the technology better, the company says that it has created a pool of 4.4 million labeled faces from 4,030 different people on its network.
Yaniv Taigman, a member of Facebook’s AI team, said that the error percentage has been reduced more than a quarter compared to the previous software doing the same task. The social network created the AI group last year to explore new ways that might help the company in deep learning. The new technology is a huge leap over the previous face-matching software and showcases a new approach towards artificial intelligence (AI) known as deep learning, an area that uses networks of simulated neurons to locate patterns in a large amount of data.
Facebook introduced a similar feature earlier
In late 2010, Facebook Inc (NASDAQ:FB) came up with a facial recognition feature that actually figures out who the person in the photo is. Initially, the feature was released to U.S. customers, but in 2011, it was released on a full scale. At that time, privacy authorities in Germany and Ireland claimed that Facebook rolled out the new feature without proper warning or consent from users. In the U.S. also, Sen. Al Franken criticized the social network for not obtaining proper consent from users. Around 350 million photos are uploaded on Facebook each day, so the feature served as a tool to speed up tagging.
The project is not entirely dependent on Facebook Inc (NASDAQ:FB), as it is still to get feedback from the research community. Facebook might have hurried up a bit in revealing the technology, as MIT’s Technology Review reports that the company released it before showcasing it at the IEEE Conference on Computer Vision and Pattern Recognition this June.