Group Emotion Perception Model Based On Facial Expression Recognition

Individual emotion analysis has a wide range of application value in the fields of medical care,education and human-computer interaction.It is an important research field in computer vision.In contrast,there are still many challenges in the research on group emotion,such as some research work ignoring contextual information at the scene level and low accuracy in identifying difficult samples.Group emotion has gradually become a favorable clue in the monitoring and management process of key public areas.In order to facilitate management departments to formulate effective group management measures,we need understand the potential emotions of groups accurately and apply them into behavior analysis solutions.Therefore,This paper mainly studies the emotions of groups in the wild and divides it into three categories for positive,neutral and negative.The main research work are as follows:(1)In order to improve the perception accuracy of group emotion,a two-channel group emotion perception model combining scene information was proposed by making full use of context information at the scene level to alleviate undetected or misplaced facial expression problems.Firstly,to solve the problem that effective feature extraction is not targeted in facial expression recognition,a Pyramid Squeeze Attention module is added in the face channel to obtain more discriminative facial features.Secondly,in order to effectively utilize emotional information other than scenes and objects,a global scene flow is developed and a multi-scale feature image fusion strategy is proposed.Finally,the features extracted from the two channels were given to the of LSTM network to learn joint distribution of features.Emotions of the group were classified into three categories of positive,negative and neutral by Softmax regression.In this paper,the accuracy of group emotion recognition in natural scene is improved to 78.40%,and the perception accuracy of global depth feature fusion is higher than that of local face feature alone,which proves the validity of the model.(2)In order to solve the problem of low recognition accuracy of the model for the difficult samples of conflicting expressions such as face occlusion,insufficient face resolution and crying and laughing,a scene perception fusion network based on the recognition of individual emotion contribution was proposed.The innovation includes three points: Firstly,considering the influence of context-related elements on the subject object,an improved attention module is proposed.It introduces the semantic and geometric features of scene space.In addition,this module can effectively identify the contribution degree of the subject object and its emotional attributes to the group emotion.Secondly,a context-aware fusion scheme is proposed to automatically adjust the fusion weight to fit the image content.Thirdly,the loss function is improved by increasing the distance range between constraint features and corresponding classes,to increase the distance between classes and narrow the differences within classes.On the GEP dataset,the accuracy of group emotion recognition in the wild is improved to 81.73%.Compared with the other four methods,the experimental results show that the proposed algorithm improves the robustness of the model and the performance of the group emotion perception.