Facial Expression Recognition: Residue Learning Using SVM

Residue learning using SVM is exploited to recognize facial expression in this paper. A facial expression consists of neutral component and expressive one(residue), which contains most of the expression information. Firstly, a cGAN is trained to generate neutral face image from an input face image. The intermediate layers record the information during this procedure. So secondly, kernel PCA and SVMs are exploited to analyze the residue in these intermediate layers. Results of experiments on five facial expression databases including BP4D, CK+, JAFFE, Oulu-CASIA and RAF show considerable performance compared with the latest methods.