Facial Expression Recognition Based On Deep Learning

Face expression is an important way for humans to express their emotions.Face expression recognition research is to automatically,efficiently,and accurately identify facial expressions.Face expression recognition is a multidisciplinary research topic such as computer vision and psychology.Under the urgency of development in fields such as security,medical care,education,and anti-terrorism criminal investigation,facial expression recognition technology has become a frontier technology with high research value.The general expression recognition method extracts the image features of a complete human face.However,different parts of the face contain different types of facial expressions.This method will extract some useless or even interference characteristics information.This information affects the accuracy of expression recognition.Based on the research results of FACS system and the actual experimental verification,this paper proposes an expression recognition method based on facial expression unit.In this paper,facial expression recognition is divided into four steps:image preprocessing,image feature extraction,face expression unit detection and expression classification.In the facial expression unit detection step,the specific part of the human face is detected and extracted using the Faster RCNN algorithm.This approach can reduce the non-critical part of the information on the identification results of interference.In the expression classification step,two classification algorithms are proposed.The first algorithm combines organ features by splicing.Then use multi-layer full-connection neural network and softmax classifier to classify expressions.The second algorithm is to construct a plurality of fully connected neural networks and a softmax classifier to classify the expressions respectively.The final result is decided by the voting results of all the expression units.This article uses the expression data set in the natural scene to experiment.The experimental results show that the recognition accuracy of this method has been improved and the speed of recognition has been significantly accelerated,which proves the superiority of the proposed method.In the method of this paper,the image does not need to be preprocessed such as alignment and complementation,which reduces the complexity of the algorithm.On this basis,a preliminary experiment was performed to address the problem of face occlusion which is most likely to affect results in facial expression recognition.Based on the experimental results,the effects of missing ratios and missing parts on face recognition were analyzed.It is proved that the method of this paper can solve the problem of occlusion expression recognition.