Facebook’s DeepFace matches faces with 97.25% accuracy
Intelligent facial-processing software
Social networking company Facebook is again expanding its empire, this time with the introduction of DeepFace - a new project which can identify whether two faces in unfamiliar photos are the same person with a 97.25 per cent accuracy rate.
The face-processing software reportedly uses a 3D model of a user's image which can be rotated so the face appears to be looking at the camera. DeepFace then uses a simulated neural network - involving 120 million parameters using locally connected layers - to figure out if there are enough similarities from the two images.
The team working on DeepFace developed the software using four million facial images belonging to 4,000 people. This will further Facebook's capabilities of suggesting users for tagging in an uploaded image. The facial verification technique has also been successfully tested with YouTube videos.