Research On Deblurring Algorithm For Face Image

Face image deblurring has been widely used in face recognition,video surveillance and other fields,and has attracted much attention from researchers.The main causes of image blur are camera shake and object movement.Image deblurring can be regarded as a deconvolution problem,which is a kind of ill-conditioned inverse problem,especially the face image deblurring is a challenging problem.Although there are many existing face image deblurring algorithms,there are still many problems that need to be solved.For example,the presence of saturated areas in the face image affects the blur kernel estimation;the significant edge extraction of the blurred face image is not accurate;the significant edge extraction of the blurred face image is time-consuming.This article conducts research on these issues,the main research work includes:(1)To solve the problem that saturated areas may exist in blurry face images,the thesis proposes a face deblurring algorithm with saturated areas(FDSA).The FDSA algorithm adaptively extracts significant edges of blurred faces,which can improve the problem of inaccurate extraction of significant edges of blurred faces.At the same time,we propose a method for detecting the saturated area of the blurred image of the face based on the blur kernel inversion.The blur kernel inversion function is used to remove the saturated area in the salient edges,so that the face deblurring is not affected by the saturated area.The experimental results show that the proposed FDSA algorithm can accurately remove the possible saturated regions in the blurred image of the face and improve the deblurring effect of the face image.(2)To solve the problem that inaccurate extraction of salient edges of faces,this thesis proposes a face deblurring algorithm based on local matching of sample sets(FDLMS).The FDLMS algorithm uses the unique structure of the face to establish the relationship between the blurred face image and the sample set.It makes full use of the sample set to improve the accuracy of the significant edge extraction of the blurred face image,and uses the extracted significant edge to guide the face image to deblur.Experimental results show that the FDLMS algorithm can provide reliable significant edges for face deblurring,and achieve better face deblurring effects.(3)To solve the problem that the time-consuming of extracting the salient edges of the blurry face image,we propose a face deblurring algorithm base on deep learning(FDDL).The FDDL algorithm proposes a face salient edge extraction network structure.The first layer of the network removes the blurred image details of the face,and the subsequent layers enhance the image edges.Compared with the FDLMS,the FDDL algorithm greatly reduces the time required to extract significant edges on the face,which can meet the real-time requirements.The experimental results show that the FDDL algorithm can improve the efficiency of extracting significant edges of the face,and achieve better face deblurring effects.