Face Unit Radial Basis Function Networks
A. Jonathan Howell · 2022
This chapter introduces a different way of learning the face recognition task through the reorganization of the standard radial basis function (RBF) networks into a group of smaller ‘face recognition units’, each trained to recognize a single person. The face unit network is essentially a normal RBF network with two output units which produces a positive signal only for the particular person it is trained to recognize. Face unit networks allow a flexible approach to learning in dynamic environments compared to other neural networks models which have to be completely retrained if the training data is altered in any way. The face unit network only uses a few of the total number of classes in a problem to train, so operations on any of the other classes not used for training will leave it unaffected. As the number of classes increases, the chance of each face unit network needing retraining due to an operation on another class will become less.