Research On Key Technologies Of Micro-expression Urecognition Based On Deep-Learning

For facial expression,it can reflect people's emotional changes and express people's emotions.For a long time,people have studied the judgment of people's emotion change through expression,especially through artificial intelligence to recognize other people's facial expression.For micro-expression,it is one of the most unconscious and real expressions,which can reflect people's current real emotions and gradually become the hot research direction of academic researchers.The change of micro-expression is very small,which makes the research of micro-expression very difficult.This kind of expression can't be forged and suppressed,so it has become an important basis for judging people's subjective emotions.Judging the change of micro-expression has a deep development in medical diagnosis,lie investigation and business negotiation.However,it is very difficult to recognize micro-expression because of the small change of micro-expression,the short time of occurrence and the appearance of micro-expression in the local area of the face.In this paper,the improvement of micro expression recognition effect is regarded as the goal,which is discussed from three aspects:One is to discuss the research status of micro expression recognition,in which it compares and analyzes the process,method and related database of micro expression recognition.The second is to propose a micro expression recognition method based on Convolutional Neural Network and Long Short-Term Memory(CNN-LSTM)feature fusion.In this algorithm,the global feature descriptor is proposed,which is mainly used in the expression of facial micro-expression.The convolutional neural network is applied to extract the spatial and temporal features respectively,and the two features are combined organically through CNN-LSTM feature fusion layer,so that the global features of the predicted micro expression can be formed.Thirdly,based on information entropy and LSTM,the recognition method of micro-expression is proposed.First,if you want to express the characteristics ofmicro-expression,you can apply optical flow image as input data.Secondly,the information entropy value of the optical flow feature image is calculated by the information entropy principle,and the information entropy value is analyzed to get the feature value,then the feature is extracted and sent to the cyclic neural network,so as to implement time sequence learning.Thirdly,the softmax classifier is applied to extract the feature vector to predict the type of micro expression.