Research On Micro-Expression Recognition And Application Based On Fusion Of Multi-scale Features

Micro-expression is a kind of facial activity with weak change and short duration.When people try to hide their true emotions,this expression will be generated unconsciously.It can reflect people's real psychological state,and micro-expression can neither be suppressed nor imitated.These characteristics make the research of micro-expression have a broad potential prospect.At present,the research on micro-expression recognition is still in its infancy,and there are many problems: 1.Compared with macro-expression,the facial movement of microexpression only exists in the local area,and the facial muscle changes are weak and the duration is short,which will lead to the loss of these subtle features in the process of feature extraction,thus limiting the recognition performance.2.Micro-expression is a dynamic process from occurrence to end,how to extract temporal information effectively and integrate it with spatial information to improve recognition performance is also urgently to be solved.Combining with the above issues,the main research of this article is summarized as follows:(1)Aiming at the dependence of micro-expression recognition on local features,a microexpression recognition method based on the fusion of local and global features is proposed.Firstly,the active region of micro-expression is divided by using the active region division strategy.Secondly,the convolution neural network is used to obtain the local and global features of each frame of micro-expression sequence.Thirdly,the local features and global features are fused to obtain the frame-level features.Finally,make full use of the physical meaning of statistical parameters in statistical methods to extract the frame-level feature vectors of the micro-expression sequence,and frame-level features of micro-expression sequences are aggregated into video-level features for micro-expression recognition.The experimental results show that the local and global fusion feature recognition results are better.(2)Regarding the problems of complex extraction process of spatio-temporal information,an end-to-end multi-channel spatio-temporal feature fusion method for micro-expression recognition is proposed.Firstly,the local and global spatial feature maps of micro-expression are extracted by parallel convolution neural network.Then,in order to extract the temporal information of micro-expression,a temporal feature extraction module is designed.The module takes the spatial feature map sequence as the input to obtain the feature-level temporal features,and obtains the spatio-temporal features by the aggregation of temporal and spatial features.Finally,the local and global spatio-temporal features are fused to get the micro-expression feature for micro-expression recognition.The experimental results show that the method greatly improves the performance of micro-expression recognition.(3)Based on the above research,the micro-expression analysis module in contactless lie detection system is designed and developed.The module has the functions of face detection,preprocessing and lie detection,and can visualize the prediction results.The module verifies the feasibility of the research method in the field of security.