Research On Facial Expression Recognition Based On Micro Expression Feature

Micro-expression has the characteristics of short duration and low intensity. It is the subconscious of human mental activity. It tends to reflect the true emotions. Based on this characteristic, study on micro-expression began to receive attention from the experts in various fields such as medicine, security, lies recognition, artificial intelligence.Based on the full study of the existing expression recognition algorithms and the difference between micro-expressions and regular expression, a method based on differential positioning and optical flow feature extraction is proposed. First, based on the theory of facial behavior encoding system, the face is divided into different facial behavior. The image sequence of facial expression sensitivity region is difference, so as to determine the movement area of the face, and to lock in the action part of the facial expression, to avoid the influence of irrelevant parts. After that. Based on the accurate positioning of facial movement region, the motion state of the pixels in each region is obtained by optical flow analysis. Principal component feature of the moving region is extracted by principal component analysis, and the dimension of the motion state of the pixel is reduced, and the motion feature extraction algorithm based on optical flow analysis and principal component analysis is proposed. At last, the main components of the optical flow of the moving regions are used as inputs, and the classification of facial motion units is judged by support vector machine. On the basis of the facial movement encoding system theory, the category of the micro expression is derived according to the corresponding relationship between the motion unit number and the characters.The key features of facial movement and the feature dimension are obtained by using the differential positioning algorithm and the main component feature extraction algorithm. Based on support vector machine, the action unit identification and expression are derived, and the difficulty of classification is reduced. Experimental results show that the algorithm has a certain increase in the recognition efficiency and accuracy, and improve the level of automatic recognition of micro expression.