Technological Research Of Micro-expression Detection And Recognition Based On Video

Micro-expressions always occur when a person tries to suppress true feelings,which differ from basic emotions because of its low intensity and short duration.Micro-expressions have a variety of application fields such as clinical diagnosis,business negotiations and interrogation benefiting from its help in conjecturing one's attempts.Consequently,it has drawn much attention from academicians for decades.Automatic micro-expression detection and recognition just start recently;the effective and robust tools has not come out on the market;most of the researchers in the field just distribute among several institutes and the necessary work in the precondition for recognition-automatic micro-expression detection seems nearly blank.The forward in micro-expression suffers from its low intensity which is easily disturbed by the environmental factors,short duration which is hardly captured by the ordinary sensors,uncertainty and appearance in a few face regions unlike the regular muscular movements of basic emotions.To solve the problems above,based on some recently proposed methods and inspired by latest success of deep learning dealing with feature learning,this paper will study and explore the method and application of micro-expression detection and recognition research.The main work is summarized as follows:(1)Propose the micro-expression detection method based on optical flow analysis and filter design.Reduce the noise by using motion feature centralization and normalization for better detection.Employ the algorithm of micro-expression recognition by regression model and group sparse spatio-temporal feature learning to distinguish micro-expression and non micro-expression.(2)Propose the deep neural network-driven feature learning method for micro-expression on sensitive facial regions based on FACS.(3)Develop an automatic detection and recognition prototype system for the micro-expressions.(4)Participate in the establishing an spontaneously induced micro-expression database and use the method of the proposed classification algorithm as the baseline evaluation.