FACE RECOGNITION USING NEOCOGNITRON NEURAL NETWORK STRUCTURE UNDER IMAGES WITH ILLUMINATION AND EXPRESSION VARIATION

摘要

This work shows one application structure of neocognitron, a convolutional neural network, in the image processing system of face recognition. During the training phase, the system uses one output neocognitron to obtain the characteristic vector of the corresponding input image. The set of the resulting characteristic vectors composes a database that will be used in the recognition phase. This type of neocognitron implementation allows its use with unlimited number of classes. As a result it was verified that in an application of 30 classes, to a number of 50 training images per subject, the recognition rate is 84.46% on patterns not used in the training phase, certifying the capacity of the network to the generalization during recognition.