A comparative analysis of different neural networks for face recognition using principal component analysis, wavelets and efficient variable learning rate

This paper proposes a new way to find the optimum learning rate that reduces the training time and increases the recognition accuracy of the back propagation neural network as well as single layer feed forward Neural Network. It involves feature extraction using principal component analysis and wavelet decomposition and then classification by creation of back propagation neural network. Paper gives a comparative analysis of performance of back propagation neural network and single layer feed forward neural network. In this approach variable learning rate is used and its superiority over constant learning rate is demonstrated. Different inner epochs for different input patterns according to their difficulty of recognition are assigned to patterns. The effect of optimum numbers of inner epochs, best variable learning rate and numbers of hidden neurons on training time and recognition accuracy are also shown. We run our algorithm for face recognition application using Coiflet wavelets, principal component analysis, neural network and demonstrate the effect of numbers of hidden neurons on training time and recognition accuracy for given numbers of input patterns. We use ORL database for all the experiments.