Facebook developing 'human-like' facial recognition - PC Retail

Facebook developing 'human-like' facial recognition

The DeepFace project is said to be 0.25 per cent shy of human facial perception accuracy
Author:
Publish date:
1-deepface-sly-web.jpg

Facebook is developing facial recognition which operates on the same level of accuracy as human beings.

The DeepFace project aims to be able to identify whether two people in side-by-side images are the same person with an accuracy of 97.5 per cent – the result achieved by human beings who have taken the same test, and 25 pe cent more accurate than existing facial recognition systems.

At the moment, the project is said to be 0.25 per cent shy of human-like accuracy, sitting at a success rate of 97.25 per cent.

The facial recognition technology used by DeepFace creates a 3D map of facial features, before flattening the model and highlighting specific features in different colours to compare to another face.

Facebook originally introduced facial recognition worldwide in 2011 as a way of speeding up photo tagging. It could be that with this new technology, no human input is required at all to tag images.

Related