A Novel Multi-stage Classifier for Face Recognition

A novel face recognition scheme based on multi-stages classifier, which includes methods of support vector machine (SVM), Eigenface, and random sample consensus (RANSAC), is proposed in this paper. The whole decision process is conducted cascade coarse-to-fine stages. The first stage adopts one-against-one-SVM (OAO-SVM) method to choose two possible classes best similar to the testing image. In the second stage, “Eigenface” method was employed to select one prototype image with the minimum distance to the testing image in each of the two classes chosen. Finally, the real class is determined by comparing the geometric similarity, as done by “RANSAC” method, between these prototype images and the testing images. This multi-stage face recognition system has been tested on Olivetti Research Laboratory (ORL) face databases, and its experimental results give evidence that the proposed approach outperforms the other approaches either based on the single classifier or multi-parallel classifier, it can even obtain a nearly 100 percent recognition accuracy.