Abstract: | Face recognition using principal component analysis (PCA) and linear discriminant analysis (LDA) suffer from the loss of accuracy when the number of classes becomes large. This paper presents an effective genetic‐based clustering algorithm (GCA) to preprocess a facial database into a two‐layer database. Then, face recognition is done to minimize the similarity criterion in a specific cluster as in the traditional PCA‐ and LDA‐based face recognition algorithms. Different from K‐means clustering, the proposed GCA introduces a novel distance and a balance factor. The distance is defined to measure the similarity effectively between a class and the centroid of each cluster, and the balance factor is designed to achieve balanced clustering results. Experimental results on the Yale‐B database in ideal and noisy conditions indicate that the proposed preprocessing method improves the recognition accuracy of the subspace recognition algorithms compared with K‐means clustering. The proposed preprocessing method is also applicable to other recognition algorithms. © 2012 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc. |