Researchers on the design and application of neural networks at Reading University have been working on a neural based signing system for deaf people to communicate using hand signing without having to be in the same room. Images of signing were filmed with the signer wearing a glove with dots on it. Then a neural net, as they are affectionately nicknamed, was trained to recognise the complex pattern that the dots formed as the hand twisted and moves and so enabling the sign to be identified. Also in the area of communications the University has been working on a video telephony project using a neural network to identify the more interesting and distinguishing features of the face, notably the eyes and mouth, for which high definiton information needs to be sent. Using a stochastic searching technique, a neural network was trained to recognise these features with 80% accuracy at identifying the eyes on previously unseen faces.
Content from our partners
