Research On Micro-expression Recognition Algorithms Based On Deep Learning

In recent years,with the rapid progress of artificial intelligence technology and the continuous development of micro-expression theory,the field of micro-expression recognition has received more and more attention.As a special kind of expression,micro-expression expresses the true feelings of human beings,so it is of great significance to study micro-expression.However,micro-expressions have the characteristics of short occurrence time and small change range,which increases the difficulty of micro-expression detection and recognition.In particular,the differences between micro-expression samples are not obvious,which makes it difficult to extract information with representative changes.The features extracted by different algorithms represent different information.Improper selection of algorithms can easily cause feature redundancy or feature loss,thereby affecting the recognition rate.Considering the features that micro-expressions need to be refined,this paper adopts the deep learning method to conduct a comprehensive and detailed research on micro-expressions.The research content is divided into three parts:micro-expression recognition based on global facial features,micro-expression recognition based on facial key points,and micro-expression recognition based on decision fusion.The micro-expression recognition model based on global facial features extracts information from the entire facial image.The model first uses the VGG face network after transfer learning as a feature extractor to extract facial image sequence information to reduce the risk of network overfitting.Then,the extracted features are input into the spatial attention module and the temporal attention module,which can focus on the regions with significant movement of micro-expression,and assign different weights to each frame of image.Finally,the adjusted micro-expression sequence features are fed into a long short-term memory network(LSTM)to process temporal features.Considering that micro-expression movements only appear in part of the face,the information extracted from the whole face may extract information unrelated to microexpressions,resulting in feature redundancy.Therefore,this paper also uses the method of face key points to identify micro-expressions.The model uses the Face mesh network in the field of face feature point annotation to extract the facial key point information of the image sequence.Then,the keypoint feature vector of the image sequence is input into the spatial attention module and the temporal attention module,and this module is used to calculate the importance of each keypoint and assign different weights to different frames.Finally,the facial sequence keypoint feature vectors adjusted by the attention module are input into a long short-term memory network(LSTM)to process temporal features.Finally,the research on the micro-expression recognition algorithm of decision fusion is carried out.The prediction output of the two models mentioned above is decision fusion,and a trainable decision maker is used to output the result.And compared with other algorithms,the results show the effectiveness of the algorithm in this paper.