Design of an optimal multi-layer neural network for eigenfaces based face recognition

Abstract

Face recognition is one of the most popular problems in the field of image analysis. In this paper, we discuss the design of an optimal multi-layer neural network for the task of face recognition. There are many issues while designing the neural network like number of nodes in input layer, output layer and hidden layer(s), setting the values of learning rate and momentum, updating of weights. Lastly, the criteria for evaluating the performance of the neural network and stopping the learning are to be decided. We discuss all these design issues in the light of the eigenfaces based face recognition. We report the effects of variations of these parameters on number of training cycles required to get optimal results. We also list the optimized values for these parameters. In our experiments, we use two face databases namely ORL and UMIST. These databases are used to construct the eigenfaces. The original faces are reconstructed using the top eigenfaces. The factors used in the reconstruction of the faces are used as the inputs to the neural network.