There are various ways to perform face recognition. Typically, a second neural network is fed with a series of pictures of people and learns how to differentiate one from another. Some algorithms explicitly map each face’s features, while others map them using more abstract features. In either case, the neural network outputs a vector for each face that uniquely identifies the person in the training set. For example, if you take a photo of a man wearing a hat, this system will identify that person filmefy .
While face recognition algorithms have improved over the last decade, there are still problems associated with this technology. A recent review by Cardiff University uncovered some shortcomings in the NEC NeoFace system. For example, the system performed worse on late nights and gloomy days. In the trial period, it flagged 2,900 possible matches. But many of these were false positives. In South Wales, this system led to 18 arrests. While Amazon’s report does not detail whether any people were actually charged, it does indicate that the technology could be used to curb fraud.
A number of developments in artificial intelligence have been made in the field of facial recognition. The emergence of deep learning technology has made it possible to build facial recognition systems that use human images and metadata to train artificial intelligence algorithms. Such algorithms can be trained on an unlimited number of images and provide feedback to improve the accuracy of the recognition. While facial recognition is still far from being foolproof, it is a promising technology that has many uses thedocweb .