Abstract:  Facial expression recognition under different lighting conditions and states is a challenging research. A fusion algorithm based on adaptive Gaussian Mixture Model (GMM) and supervised super-vector encoding is proposed. Firstly, the overlapping image blocks are extracted. Then, local descriptor from each block is extracted by the adaptive GMM so as to map images in low-dimensional space to high-dimensional space. Finally, supervised super-vector encoding is used to do classification training. Experimental results on the Multi-PIE and BU3D-FE multi-view facial expression databases show that the recognition accuracy of proposed algorithm can achieve 91.8% and 95.6% respectively on Multi-PIE and BU3D-FE. It takes only 0.142 seconds in identifying a sample on BU3D-FE. Proposed algorithm has higher recognition accuracy and less recognition time-consuming than several other excellent algorithms.

Key words: facial expression recognition, adaptive, Gaussian Mixture Model(GMM), supervised learning, super-vector encoding