A sequential subspace face recognition framework using genetic-based clustering

Different from other classification problems, there are usually a large number of classes in the face recognition. As a result, the recognition accuracy of the traditional subspace face recognition algorithm is unsatisfactory. This paper presents a sequential subspace face recognition framework using an effective genetic-based clustering algorithm (GCA). Firstly, the facial database is decomposed into a double layer database using a face recognition oriented GCA. Then, the face recognition is realized by minimizing the distance measures in a specific cluster as in the traditional subspace face recognition algorithms. The contributions of this study are summarized as follows: 1) The class, i.e., person is regarded as an element in the clustering rather than an image. 2) The proposed GCA uses a novel distance to measure the similarity between a class and the cluster centroids of different clusters. 3) The proposed GCA uses a balance factor to achieve balanced clustering results. Experi mental results on the extended Yale-B database indicate that the proposed sequential subspace face recognition framework has higher accuracy compared with the traditional subspace methods and K-mean+traditional subspace methods.