Research On Spontaneous Facial Expression Recognition With Multi Head Pose Based On Deep Features

Facial expression recognition technology gives computers the ability to understand human emotions and intentions,and it has important application value in human-computer interaction,learning emotional analysis and psychological treatment.In these practical applications,the user's head pose is unconstrained.However,most of the current research work on facial expression recognition is only aimed at positive face images,which cannot meet the needs of many practical applications.To promote the low recognition rate caused by the head pose in the natural scene,this paper proposed a conditional probability model on the spontaneous facial expression recognition with multi head pose.Specific research includes:(1)A head pose estimation algorithm based on deep feature is proposed.The head pose is expressed by deep features,which is derived from the full connection layer of the deep model,and is obtained through the forward propagation calculation of the deep model.Different head pose classifier is obtained by the training deep features of support vector machine,random forest and neural network.Finally,the performance of each classifier is judged by testing.The test results show that the classification ability of random forests is the strongest based on the same deep features.At the same time,compared with the existing algorithm,the experimental results show that the method used in this paper is more accurate for head pose estimation.(2)A facial expression recognition algorithm based on conditional probability model is proposed.Based on the proposed head pose estimation method,the conditional probability model is set up.The estimated head pose is used as the conditional probability,and the expression classifier under different head pose is trained respectively.The final result is estimated by using the Bayesian formula to synthesize the output of the expression classifier under different head pose.A large number of comparative experimental results show that the expression algorithm based on conditional probability model is better than the existing algorithm.The probability model of head pose condition is better than that of non-head pose condition,which proves the effectiveness of the method.