Modular neural networks for face detection

A new concept for detection of human faces is presented. An efficient approach to reduce the computation time taken by neural networks for the searching process is introduced. We combine both Fourier and wavelet transforms with cooperative modular neural networks (MNNs) to enhance the performance of the detection process. Such an approach is applied to identify human faces automatically in cluttered scenes. Here, neural networks are used to test whether a window of 20×20 pixels contains a face or not. The major difficulty in the learning process comes from the large database required for face/nonface images. A simple design for cooperative MNNs is presented to solve this problem by dividing these data into some groups. Such division results in reduction of computational complexity and thus decreasing the time and memory needed during the test of an image. In order to have a faster detection algorithm, a combination of the FFT and the wavelet transform is made in order to reduce the elapsed time during the test phase and enhance the detection performance. Feature measurements of the input faces are made through Fourier descriptors which are insensitive to rotation, translation and scaling. Such a feature is modified to reduce the number of neurons in the hidden layer. The second stage extracts wavelet coefficients that have been shown to provide advantages in terms of better representation for a given data to be compressed. Finally, the resulting vector is fed to one of five neural networks for face detection. Compared to previous work in face detection, the use of this combination reduces the number of neurons required for neural networks. Simulation results for the proposed algorithm show good performance on detecting faces with rotation, occlusion, noise, or change in illumination