Block-based Independent Component Analysis and Face Recognition
基于分块独立分量分析的人脸识别

A novel feature extraction method using block-based independent component analysis (BICA) is proposed in this paper. BICA partitions the facial image into a few blocks, reducing the influence of some factors such as lighting condition and facial expression on face recognition. The method takes the row and the column vector of the reconstructed matrix as the training vector sequentially to extract independent components. Since the dimensionality of the training vector in Block-ICA is much smaller than that in the traditional ICA, it can reduce the face recognition error caused by the dilemma in traditional ICA, i.e. the number of available training samples are great less than thoes of the training vector, and thus reduce the recognition time. Experiments on the Yale and AR databases validate the effectiveness of the proposed method.