Micro-expression Recognition Based On Convolutional Neural Network And Its Effectiveness Research

With the fast and furious development of internet technology,it has brought AI,big data analysis and emotion analysis,and it has received much more attention.Emotional recognition,which is attached to emotion analysis,has become a scorching research issue in recent years,which benefits people in real life and all areas of society.Nonetheless,facial expression can be divided into macro-expression and micro-expression according to dynamic range,time length and inducing factors.Nowadays,most facial expression recognition is classified into macroexpression.But the micro-expression movement is subtle and the duration is relatively short,so the recognition of micro-expressions was mainly carried out by manual recognition through training of professionals.With the rapid development of computer vision technology,spontaneous micro-expression datasets appeared,which made the micro-expression recognition algorithm enter the rapid development stage.Micro facial expression is an irrepressible subconscious expression that reflects the true feelings of the human heart.Therefore,most applications of micro facial recognition are in the fields of medicine,interrogation,security and negotiation.However,based on the characteristics of micro-expression,neither the manual recognition of professionals nor the existing computer vision algorithms can recognize the emotion categories of micro-expression accurately.Currently,the bottleneck and difficulty of micro-expression recognition are information redundancy,insufficient sample size and difficulty in detailed feature extraction.Therefore,according to these issues,this paper aims to improve the accuracy and generalization ability of micro-expression recognition,based on convolution neural network,carries out research on micro-expression recognition methods,including the following contents:For the redundancy of information and insufficient sample size,this paper first carries out the preprocessing work,effectively removes the redundant information of the image sequence of the dataset by face cropping,peak frame positioning and double frame extraction,maximizes the dynamic information,and then expands the sample data by mirror flip and random rotation to create the database before recognition.In order to overcome the obstacle of detailed feature information extraction,this paper combines dense optical flow method and convolution neural network to construct an end-to-end recognition algorithm,OFCNN.Which captures detailed time features with dense optical flow,and the extraction of spatial features is done by lightweight convolution neural network,that ensures the full extraction of dynamic information of micro-expression,and realizes the recognition and classification of micro-expression.Then,the performance of this method is tested by CASME dataset and CASMEII dataset.The micro-expression recognition method proposed in this paper is evaluated from many angles,and the accuracy rate and F1 value are selected as the evaluation criteria.Firstly,the method is compared with the existing manual method and depth method,and the performance of this method is proved to be better.Secondly,the method is evaluated by using the hybrid dataset,and the generalization of this method is proved.