Face Image Motion Blur Removal Algorithm Based On Generative Adversarial Network

In recent years,face image is widely used in video monitoring and recognition,and face image data is more and more important.In the process of obtaining face image,fuzzy is often produced,the most common is the relative motion between camera and character,called motion blur.The face image with motion blur will greatly affect the effect of monitoring and recognition.Therefore,it is of great significance to remove the motion blur of face image in computer vision task.In the field of computer vision,the existing face image de fuzzy algorithm does not aim at the problem of motion blur,but the existing algorithm of removing motion blur does not target face image.Therefore,the two algorithms are not effective in the face image motion fuzzy scene.In this thesis,we propose a motion blur algorithm for face image removal,aiming at the problems existing in the two algorithms.The main research work and contribution are as follows:(1)This thesis presents a motion blur algorithm for face image removal.The whole structure of the algorithm is to generate the confrontation network,which is divided into the backbone network and the bypass network.The face analysis information is integrated into the bypass network,which enriches the characteristics of the network and effectively removes the motion blur in the face image.(2)In this thesis,the self-coding network is integrated into the neural network model as a preprocessing module to improve the effect of the whole network to remove motion blur.This thesis does not preprocess the image,but integrates the self-coding network as a preprocessing module into the whole network,and obtains an end-to-end network model.(3)In this thesis,face analysis graph is integrated into the network as prior information,and it is added to the final loss function as a part of the loss.This thesis designs a bypass network structure to generate face analysis information,and integrates it into the backbone network,enriches the image feature information in the network,and helps to locate the edges of face images.In this thesis,the loss function of face analysis graph is added,which can effectively train for bypass network and improve the quality of network output image.(4)This thesis makes a data set for the fuzzy motion of face images,and provides 11 modules of all face images.Firstly,through artificial screening,face images meeting the requirements are selected in the open face data set.The motion blur algorithm is used to blur these face images randomly.Then,the clear face image and the face motion blur image are combined into image pairs,and then divided into training set and test set to obtain the face image motion fuzzy data set;finally,the face image motion fuzzy data set is obtained;finally,the face image motion fuzzy data set is obtained by making the face images move fuzzy The clear face image is divided into 11 modules by image segmentation algorithm to get the face analysis image.The experiment shows that the method can effectively remove the motion blur of face image and has good effect on visual perception.At the same time,compared with the most advanced algorithms,the algorithm has obtained the best performance in two different image quality evaluation indexes.