Sparse representation and random forests based face recognition with single sample per person

Traditional face recognition methods usually require a large number of training samples. In some specific applications, however, we can only obtain one facial image as training sample for each person, which is usually referred to as single sample per person face recognition. The recognition rates will decrease dramatically using traditional methods in such situations, and some may even fail to work. To address this problem, we propose in this paper a novel face recognition approach based on sparse representation and random forests. We first divide each face image into multiple patches. And then we employ sparse coding to obtain local image features and random forests to acquire global features. Finally, we use L1 based nearest neighbor classifier to identify the unknown face image. Experiments are carried on two widely used face databases AR and FERET. The experimental results demonstrate our proposed approach is effective and promising.