MICRO-EXPRESSION RECOGNITION BASED ON THE SPATIO-TEMPORAL FEATURE

Micro-expressions are brief and involuntary facial movements which reveal persons' real emotions. Recognition of micro-expression is a great challenge due to its properties of short duration and low intensity. To address this problem, we propose a ROI (Region of Interest)-based spatio-temporal feature named Dense Sampling Optical-flow's Mean Magnitude and Angle (DS-OMMA) for micro-expression recognition. Namely, partitioning the facial region into some adaptive ROIs discovers the facial spatial structure, and optical flow explores the temporal information by capturing small muscular movements on the face. Moreover, dense sampling reduces the effect of noise caused by head movement or illumination. The proposed approach is evaluated on two spontaneous micro-expression datasets, i.e., CASME2 and CAS(ME)(2). The experimental results show that our proposed DS-OMMA feature performs better than the baseline feature LBP-TOP and the state-of-the-art feature MDMO in recognition accuracy.