Research On Multi-view Facial Expression Recognition Based On Multi-feature

With the advent of the artificial intelligence era,more and more attention has been paid to human-computer interaction,and the improvement of multi-view facial expression recognition technology will make human-computer interaction a qualitative leap.At present,multi-view facial expression recognition technology has been applied in many areas of life.The facial expression recognition rate of frontal faces is relatively high,but most of the faces captured by the camera are not frontal faces.Therefore,the improvement of multi-view people facial expression recognition rate is particularly important.This paper will study the three aspects of multi-view facial expression recognition,which are image preprocessing,feature extraction,classification and recognition.The main work is as follows:1?The preprocessing includes face segmentation and nose segmentation.Pure face segmentation is the basis of facial expression recognition,and nose segmentation is important for facial head pose discrimination.In this paper,the YCrCb color threshold segmentation model is used to extract the pure face image,and the convolutional neural network is used to realize the head pose discrimination of the face.The experimental results show that it has good segmentation effects for different face orientations,and the convolutional neural network method based on nose image feature information in this paper has a good face head pose recognition effect.2?Feature extraction includes Gaussian Markov random field texture features of wavelet transform domain signals and Histogram of Oriented Gradient shape feature.Wavelet transform has the advantage of signal multi-scale analysis.Therefore,the Gauss Markov random field features extracted from the wavelet transform signal of the face image can more clearly express the difference between the classes of facial expressions.The Histogram of Oriented Gradient method can completely calculate the gradient amplitude and direction of the pixels in the local area through the scanning of a fixed window and mathematical operations,and then effectively extract the shape features of facial expressions.The experimental results show that the combination of wavelet random field texture features and histogram of Oriented Gradient shape features proposed in this paper can effectively improve the recognition rate of multi-view facial expressions.3?The classification and recognition use the support vector machine algorithm,and the formation of the feature vector is the key to the recognition rate.This paper studies the effect of feature block and block combination on the classification results.A reasonable number of blocks and a reasonable combination of different blocks will improve the recognition rate of multi-view facial expressions.This article gives the best block the combination of quantity and block.In this paper,we carry out experiments on KDEF dataset with five different angles of facial expression recognition and compare them with other methods.The effectiveness of the multi-feature and multi-perspective expression recognition algorithm is verified.The experimental results show that it has good recognition effect under five different angles,and the average face recognition rate of different angles is 96%.