Research On Wild Facial Expression Recognition

Facial expression recognition is the feature extraction and facial expression images classification using of pattern recognition and other computer technologies,so that computers can understand human emotions.Facial expression recognition has a wide range of applications in human-computer interaction,medical treatment,education and other fields,and has become one of the hot research directions in the field of computer vision.At present,the facial expression recognition technology in the laboratory environment has been extensively studied.However,the facial expression recognition in the wild is more suitable for practical applications.But there are many problems,and the performance of facial expression recognition algorithms also needs to be further improved.This paper mainly studies the facial expression recognition algorithms in the wild.The main research contents are as follows:(1)In order to solve the problems of imbalanced data for wild facial expression recognition and large intra-class variation,this paper proposes a class-balanced and local median model(CALM)for wild facial expression recognition algorithm.The CALM model is divided into three modules: the preprocessing module,the class-balanced and local median loss jointly supervised deep learning network module and the output module.Its innovation is a new loss function called CALM Loss(Class-balanced and Local Median Loss)is proposed.The CALM Loss includes two parts: the class-balanced softmax loss function and the local median loss function.The class-balanced softmax loss function marks the two expressions of fear and disgust,which have a small amount of data and are prone to misclassification,as difficult samples,and the other five expressions as easy samples.During the network training,the weight of difficult samples is adaptively increased to improve the recognition accuracy of difficult samples.The local median loss function uses the median value of several neighbors that belong to the same category as each sample as the class center,which can reduce the impact of outlier samples on the choice of category center to a certain extent.A large number of experiments on RAF data set have proved that the CALM model proposed in this paper can improve the accuracy of facial expression recognition in the wild to 85.38%,and can solve the problems of data imbalance and large intra-class variation at the same time.(2)In order to make the model pay more attention to the area related to the expression and obtain more discriminative expression features,this paper proposes a dual-channel based on the attention mechanism of facial expression recognition model(DPAM).The DPAM model is improved on the basis of the CALM model,the single-channel network is changed to the dual-channel network.The dual-channel network contains the entire face area channel network and the partial face area channel network.The entire face area channel network is used to extract features of the entire face image,and the partial face area channel network is used to extract features of the local face.At the same time,an attention mechanism module is added to the network structure.The attention mechanism module consists of two parts,namely what pathway attention model and where pathway attention model.What pathway attention model is used to learn the importance of feature channels,and where pathway attention model is used to learn the importance of regions in the image.The experimental results show that the DPAM model can obtain more discriminative features and improve the recognition accuracy of facial expression recognition in the wild to 85.87%.