Facial Image Deblurring

Image deblurring is widely used in real life,such as video monitoring,medical imaging and so on.Facial image deblurring,as an important part of image deblurring,has been developed for a long time.But the state-of-the-art deblurring methods failed in getting the best result.Existing deblurring methods mainly focused on extracting explicit or implicit face edges to estimate the blur kernel,while there is no obvious texture and edges in the face image.Existing deblurring methods directly or indirectly utilized image salient edges to estimate the blur kernel rather than designing corresponding image feature for a particular type of image.In order to take full advantage of the structural information of human face image,Pan et al.proposed a method of setting up a standard library of face contour images.But the method of manually extracting contour leaded to inaccurate contour extraction.Meanwhile,because the standard library was mainly composed of front face images,the effect of complex human face image processing was poor.This paper is based on the feature extraction of face contour.This feature can fully reflect the fuzzy degree of the blurred face image which makes the blurred image contour extraction more accurate.Experiments show that the proposed method can better extract the structure of the human face and deal with the complex face image.On the basis of the above method,in order to improve the complexity produced by the operation of setting up the face database and the candidate image matching,this paper deeply studies the alignment method for the accurate location of the face contour,together with L₀-norm deblurring method.In this paper,a deblurring method based on face alignment,is proposed.The core is based on the regression of the forest to extract the face contour.The extracted face contour plays the same role as the salient edges of the natural image,and then guides the deblurring process.The results show that the proposed method can not only deal with the complex face images,but also greatly reduce the computational complexity.