Facial Micro-expression Recognition Based On Monogenic Binary Pattern

Micro-expression is a brief and involuntary facial expression which only can be detected by a few of people in real life. Compared with normal facial expression, micro-expression is more likely to reveal people’s feeling and motivation. It usually appears when people try to conceal or repress their real emotions, so micro-expression is an effective clue for lie indication. As for its short duration and low intensity, the research of automatic micro-expression detection and recognition is a challenging task.This paper presents a method for micro-expression feature extraction base on monogenic local binary pattern, and applies it to detect micro-expression in static images and dynamic image sequences. The main work is as follows:(1) Facial landmark location. As the computation in parameterized appearance models(e.g., Active Appearance Model) is expensive and complexity, this paper presents the supervised descent method(SDM). SDM extract features by learning the descent directions of the sequences in a supervised manner, which reduces the calculation..(2) This paper presents the monogenic binary pattern(MBP) for micro-expression feature extraction. This method uses fewer convolutions to extract more compact feature vectors, which reduces time and space complexity. Dynamic feature can represent more comprehensive information of micro-expression, so MBP is expanded to three orthogonal planes MBP to classify better.(3) As probability statistics models, Hidden Markov Models has been applied in speech recognition and facial expression recognition successfully. A method based on the Hidden Markov Models is presented here which gains a HMM for each micro-expression.The experiments simulate by MATLAB. There are two micro-expression image databases used in the experiments, CASME and SMIC. Experimental results show that the proposed method has better performance than Gabor and reduce the complexity in time and space.