Show most any adult human two different photographs of another person and they'll be able to match them with nearly 100% accuracy. These are the kind of results Facebook's latest research is also zeroing in on.
Technology Review reported this week that Facebook has developed new facial verification software capable of matching the same person correctly nearly every time, regardless of whether the subject is actually facing the camera.
Using new artificial intelligence technology known as "deep learning," researchers working for the social network have managed to reach a 97.25% accuracy, even with images where the lightning is different.
Those results compare favorably to human beings, who are said to match two different images of the same person with a typical 97.53% degree of accuracy.
"You normally don't see that sort of improvement. We closely approach human performance," explains Facebook AI team member Yaniv Taigman, noting the latest software has eliminated nearly one quarter of facial matching errors found in earlier versions.
DeepFace uses "facial verification" software to match two different images where the same face appears, rather than focus on the relatively easier task of only recognizing a person based on their facial characteristics.
The new software accomplishes this task in two steps: First by correcting the angle to make the subject face forward, then by applying deep learning from a "simulated neural network" to a 3D model until the software is satisfied both images are from the same face.
Facebook launched its research group late last year to help increase the accuracy of faces tagged on the social network. One day, we could see this face-spotting tech show up in features such as automatically suggesting tags whenever users upload new photos.
- Is it "game on" for Xbox One? Find out in our full review!