Region Based Fuzzy Neural Networks for Face Detection

Abstract

Proposes a fuzzy neural network method for face detection. In the proposed method, fuzzy membership degrees are assigned to preprocessed 20/spl times/20 window face and non-face image regions. These fuzzy membership degrees are then input to a neural network to be trained using the error backpropagation training method. After training, the output value of the neural network is interpreted as the degree of which a given window is a face or nonface region. If the window is determined to contain a face, post-processing is then performed. Experimental results show that the proposed method can detect face images more accurately than using conventional neural networks. Also, the proposed fuzzy neural network architecture is shown to require less hidden neurons than when using conventional neural networks.