Dynamic Facial Expression Recognition based on the Fusion of Handcraft Features

Facial expression recognition becomes an attractive research area in the field of biological recognition, which plays a vital role in a wide range of the application of the medical treatment, security systems, education, and so on. Extracting discriminative facial expression features has been an essential solution to improve the performance in dynamic facial expression recognition tasks. In this paper, we present a novel dynamic facial expression recognition framework that fusing geometric features with semantic features to effectively extract robust features. First, the proposed distance difference method is used to identify the peak frame in an image sequence, and the geometric features are calculated by the proposed geometric feature-based method. Then the geometric features are described semantically to construct a semantic feature set. Finally, we conduct an assessment on these geometric features according to the feature intensity standards and semantic rules to obtain the final fused features. Our method is carried out on two benchmark datasets: CK + (Extended Cohn-Kanade) and MMI (M & M Initiative). Its efficiency is demonstrated by extensive experiments. The best result achieves 98.73% recognition rate as classifying the 6 classes on CK+ and 87.36% on MMI, outperforming the results of state-of-the-arts.