Use of cluster validity in designing adaptive Gabor wavelet based face recognition

Abstract

Face images in various situations due to facial expression, view point, illumination conditions, noise, etc. make identification process difficult. In this paper, the situation information of face images, what we call image context, is used to improve performance of a face recognition system. The proposed system partitions face images into several image contexts (groups) based on cluster validity, and takes adaptation to individual partitioned groups. In Gabor wavelet based face recognition, we apply weights to individual elements of facial feature, and those weights are trained by Genetic algorithm. We tried to use several unsupervised learning methods, clustering algorithms here, to partition face images into proper image contexts. There exists no formal way to decide the suitability of clustering algorithms for aiming at high recognition rate. We discuss about the process of cluster evaluation using the proposed cluster validity measure in designing adaptive face recognition. We achieved encouraging results though extensive experiments. © Springer-Verlag Berlin Heidelberg 2006.