Research On Automatic Recognition Algorithm Of Micro-expressions Based On Video Sequence

Facial expression is one of the important ways to express emotion.However,the fa cial expression has a high camouflage.And people deliberately show a conscious expres sion,compared to the most realistic feelings and motivations can be more reflected thro ugh the micro-expression.Microexpression is a dynamic process,so time and space dim ensions contain a large number of micro-expression-related information.Almost all stud ies so far have remained in the analysis of micro-facial space characteristics,but ignored the micro-expression on the time dimension of the information.In this paper,we use the video sequences to automatically identify the micro-expression of the target.In the mean time,we research micro-expression dynamic feature extraction,feature dimension reduction and classifier design.Aiming at the characteristic of short duration,low intensity and easy to be neglected by the naked eye,this paper proposes an automatic micro-expression recognition algorithm.Since the duration of each kind of micro expression is different,this paper proposes a preprocessing technique based on the frame interpolation and inter frame motion compensation.This technique solves the problem of time series normalization,which is convenient to describe the video sequence.We successful employ histogram of 3D gradient descriptor to extract micro expression information from video sequence.This feature can deepen our understanding of the dynamic model of micro expression.The feature extraction method can extract the dynamic micro expression information from the image sequences,however,the dynamic features are as high as several thousand.In the face of "Curse of dimensionality",this paper proposes a weighted k-nearest neighbor classification algorithm based on principal component analysis and ReliefF feature.This method not only reduce the feature dimension but also improve the accuracy of micro expression recognition.In order to further improve the accuracy of micro expression recognition,this paper we put forward an adaptive weight calculation method based on block modeling and block fuzzy classification.The proposed method can effectively reduce the sample error rate and increase the robustness of the test samples.The experiment results show that the recognition accuracy rate of histogram of 3D gradient descriptor is 22.05% higher than that of state-of-art LPQ-TOP.And combining principal component analysis and ReliefF feature weighted k-nearest neighbor classification algorithm is proposed to reduce the feature dimension,and to some extent,it can effectively improve the accuracy of the micro expression recognition.And the micro-expression recognition accuracy of adaptive weight calculation method based on block modeling and block fuzzy classification method is 9% higher than the accuracy of histogram of 3D gradient descriptor.