Micro-expression Recognition Based On Video

In recent years,micro-expression has attracted more and more attention because it is not controlled by humans but it can reflect the most realistic psychological activities of human beings.The duration of the micro-expression is very short and its expression is small and it is hard to be noticed by people,but it contains rich emotional information.Micro-expressions are more and more known by people and more and more interested in the mysteries contained in micro-expressions.More and more researchers in various fields have carried out cross-professional cooperation,which has promoted the rapid development of micro-expression research,and the research work has gradually deepened.In the process of micro-expression recognition,the quality of the extracted features plays a decisive role in the accuracy of the recognition.The commonly used feature extraction methods are 3D histogram,optical flow method,LBP-TOP,etc.LBP-TOP algorithm works well in video-based micro-expression recognition,but because LBP algorithm compares the size of the center pixel and the neighborhood pixel,it is not sensitive to light and easy affected by noise.Therefore,this paper has mainly done the following work on the basis of the above:(1)Two methods of face detection are introduced.Because the motion of the micro-expression is small,the face position and posture of each person in the video are different.The method of normalizing the face geometry is used to manually determine the position and size of the human face by selecting the positions of the eyes of the person's eyes,and the face image is cropped.(2)A micro-expression recognition method based on optical flow method and spatio-temporal local binary pattern is proposed.By performing optical flow operations on the facial micro-expression video sequence images,horizontal integral projection and vertical integral projection are performed,and two feature histograms are obtained through LBP encoding,and the appearance model features are cascaded.The horizontal or vertical integral projections of all the images are used as new texture images,and the horizontal or verticalmotion feature histogram is obtained by LBP coding,and these feature histograms obtained by cascading are obtained.Using the characteristics of the resulting micro-expression,training and classification are performed by a support vector machine.(3)A method of micro-expression recognition based on deep learning was proposed.Combining LBP with CNN,the first convolutional layer of the convolutional neural network was modified to encode the convolution filter with LBP,and the activation function calculation and weighted fusion were used as the next layer input of LBCNN.Then use neural networks for training and classification.