Learning from essential facial parts and local features for automatic facial expression recognition

In this paper, we develop an automatic facial expression recognition system which establishes relations between facial expressions and the facial parts changes. Here, the differences between neutral and emotional states are used to help locating and identifying the essential facial parts for human expressions. For face description, region-based method to compute LBP features is applied then the most important ones for each expression are selected. As the system combines LBP and Gabor features, it can recognize the facial expressions efficiently. The method is evaluated on JAFFE and Cohen-Kanade database and it performs better and is more stable than other automatic or manual annotated systems.