Skip to Main Content
A modified independent component analysis method, termed the matrix-based modular independent component analysis (MMICA), is developed in this paper. The main idea of the proposed method is that each of all facial images is first partitioned into many subimages. Every subimage is regarded as a new training sample, by which a new set of training samples is formed. Since the dimensionality of each of the subimages is much smaller than that of each of the original training images employed in traditional ICA, it can reduce the face recognition error resulted from the dilemma in ICA, that is, the small sample size problem (SSS). Then, the proposed algorithm performs the whitening step directly based on the two-dimensional subimages, which can improve the efficiency of the proposed method. Experimental results on the Yale and AR databases show that the MMICA method outperforms the traditional ICA and PCA methods.