Micro-expression Recognition Based On Attention Mechanism

Micro expression is a kind of facial expression which is fast,hard to observe with eyes and can reflect the real feelings of human beings.Due to micro expression can reflect people's real feelings and motivation,it is often used in business negotiation,police diagnosis,psychoanalysis and other fields.However,due to the characteristics of short duration(1/25-1/2 seconds)and subtle changes(small muscle contraction or relaxation),micro expression recognition is still a huge challenge.Deep learning is especially popular in recent years,and the field of neural network and attention mechanism is more and more extensive,which provides a new research direction for micro expression recognition.Therefore,this paper uses deep learning to create an automatic micro expression recognition algorithm to select more representative micro expression features for micro expression recognition,and implements a micro expression recognition system.This thesis studies the following aspects(1)In this thesis,a method of micro expression recognition based on 3D2 D hybrid structure is proposed.Firstly,we used multiple three-dimensional convolution layers to extract spatiotemporal features,and then two-dimensional convolution and attention mechanism are used to extract higher-level and more critical spatial features.Finally,Softmax layer is used to classify micro-expressions.A series of experiments on micro expression dataset show that the model is better than other methods.(2)In this thesis,a three-dimensional convolution micro-expression recognition method based on Attention mechanism is proposed.This method is a dual-flow structure,which can effectively use the features of image sequence and optical flow sequence.The Attention layer and Co-Attention layer are used to extract more effective micro-expression features to better solve the micro-expression recognition problem.Experiments on the micro expression dataset show that the proposed model has better recognition effect.(3)In this paper,the first automatic micro-expression recognition system based on attention mechanism-AMMRS(Attention Mechanism for Micro-Expression Recognition System)is built,and the micro-expression recognition algorithm proposed in this paper is applied to the micro-expression recognition function module of this system,and complete the development and testing of the whole system...