Facial Expression Sequence Interception Based on Feature Point Movement

Automatic and accurate Facial Expression Sequence Interception (FESI) for the face video is an enormous challenge to Facial Expression Recognition (FER). This paper proposes a framework for FESI by the relative movement between facial feature points. Firstly, Active Appearance Model (AAM) is used to mark the feature points and the most representative 24 of them are chosen. Secondly, the intraframe Euclidean distances between any two feature points are calculated and the interframe Euclidean distances variation polylines are fitted, thus the slopes of different segments in the polylines are obtained. Finally, facial expression sequence is intercepted from the face video by determining two key frames whose emotional intensities are minimum and maximum, respectively. Corresponding experiments on Beihang University (BHU) facial expression database and MMI facial expression database are conducted whose results have shown the satisfactory interception effect of the proposed method.