Research On Dynamic Expression Recognition Of Occluded Faces

In recent years,with the development of smart cities,the demand for humancomputer interaction is getting higher and higher,and facial expression recognition provides the possibility to realize human-computer interaction.However,in the actual face detection scene,the collected dynamic face samples There is occlusion,and the accuracy of facial expression recognition is not ideal.Therefore,this paper studies the occlusion problem of dynamic facial expressions in real and complex scenes.The main research contents are as follows:This paper proposes targeted solutions to the problem of face occlusion in dynamic facial expression recognition.An occluded facial expression recognition model is constructed,based on the convolutional neural network VGG16,an occlusion determination unit is designed to strengthen the extraction of expression features in unoccluded areas or areas with less occlusion;combined with residual network to obtain full face features,to prevent the loss of information in the occluded area.This model can effectively improve the accuracy of facial expression recognition for occluded faces.Aiming at the segmentation of dynamic expression sequences and the acquisition method of key frames,this paper proposes a new dynamic sequence segmentation rule based on SSIM algorithm,which is used to segment dynamic expression sequences and extract the key frames of each expression sequence;And established a dynamic expression sequence database,through the comparison of manual annotation and experimental results,the proposed algorithm rules were experimentally tested,and finally satisfactory results were obtained.In the process of recognizing dynamic facial expressions,this paper pays full attention to the time series information in dynamic facial expressions,uses GRU network to extract time series information,and integrates it with the occlusion perception system to improve the dynamic expression recognition system for occluded expression recognition.The accuracy rate makes it more applicable in real scenarios.