Facial Expression Recognition by SVM-based Two-stage Classifier on Gabor Features

We propose a two-stage classifier for the elastic bunch graph matching based recognition of facial expressions. The major purpose is to calculate distinctive similarity between image patterns by applying optimal weights to responses from different Gabor kernels and those from different fiducial points. In the first stage, we perform SVM on each fiducial point individually to extract a weighted feature from the Gabor response. The optimal fusion of those features is then calculated by another stage of SVM, providing the weight between fiducial points. From numerical experiments, the proposed method shows improved performances when comparing with other methods.