Facial Expression Recognition Based On Multi-feature Fusion

With the rapid development of computer vision technology,people are increasingly eager for a more intelligent human-computer interaction experience.As an important bridge for human-computer intelligent interaction,facial expression recognition technology has become a current research hotspot.Expression recognition technology has been applied and developed in many real-life fields,such as driver status monitoring in the transportation field,patient monitoring in the medical field,etc.But the recognition rate problem limits facial expression recognition technology in more fields in the application.How to extract the features of facial expression images more effectively and further improve the characterization of facial images requires more effort.Aiming at the current problems of facial expression recognition,this thesis mainly studies the image segmentation,feature extraction algorithm and facial expression recognition in facial expression recognition.The main work of this article is as follows:The full convolutional network is used to complete the facial mouth area segmentation,which improves the mouth segmentation accuracy.Mouth segmentation is the basis of facial recognition using shape features.After experimental verification,this method has a good effect on the facial segmentation of facial expression images.It can be accurate under different lighting backgrounds,different facial skin colors and facial shapes split the face mouth.Two improvement methods for Gaussian markov random field model are proposed,which solves the problem that the texture information obtained by Gaussian markov random field feature in face expression recognition is not fine enough,and improves the ability of feature character character image information.The first is a Gaussian Markov Random Field(BPGMRF)feature model fused with a binary model.The model algorithm not only maintains the advantage of the strong spatial expression of the Gaussian Markov random field model,but also can express the size relationship of the pixel gray value of the local area,as well as the local pixel gray value and the overall average gray value.Size relationship.The second improved algorithm is a fusion of Gaussian Markov Random Field(MP-GMRF)feature model with maximum pooling.The algorithm adds maximum pooling to increase the robustness of features and the ability to characterize image information.An edge entropy shape feature extraction algorithm at a specific angle is proposed to improve the representation ability of facial expression information..The mouth is the organ in the face area that changes the most with the change of expression.The entropy calculation is performed by the vertical distance between the center point of the mouth and the edge point,and the shape change of the edge of the mouth is transformed into certainty.The combination of texture features and shape features further enhances the representation ability of expression image information and improves the accuracy of expression recognition.Experiment on JAFFE dataset with seven facial expressions.Compared with the methods proposed in the literature published in recent years,the results show the effectiveness of this method in facial expression recognition.