ZMI and wavelet transform features and SVM classifier in the optimized face recognition system

This paper compares performances of the Zernike moment invariant (ZMI) and wavelet transform features in the application of face recognition. In this study, after preprocessing and face localization of an image, we optimize the exact location of oval shape of face in the image with genetic algorithm which improves the recognition rate. High order ZMI and discrete wavelet transform (Haar wavelet) is utilized to produce feature vectors. In the wavelet transform step, we used Mallat pyramid algorithm for finding approximation of the image in lower resolution and decomposed each image in 4 resolution level. Also SVMs classifier which is a new learning machine and has very good generalization ability has been used as a classifier with two different kernel functions. Simulation results on ORL database show that approximately the same results are obtained for both ZMI and wavelet features. But feature extraction using wavelet transform has a rate of 0.078 image/sec that is about 11 times faster than the rate of ZMI feature