Blur Image Assessment And Its Application In Video Deblurring

As one of the main ways of information transmission,images and videos may introduce different types of distortions during collecting and transmitting.Distortion affects the quality of images and videos.Blur is one of the most common types of distortion.First,an effective blur image quality assessment method is proposed in this paper.Then,the blur image quality assessment method is applied to video deblurring.Finally,an effective video deblurring method is proposed.The main contributions of this paper are as follows:(1)The previous methods cannot show good performance simultaneously on synthetic blur databases and real blur databases.In this paper,a new blur image quality assessment method based on response function of singular values(RFSV)is proposed,which performs well on both classes of databases.The proposed method consists of four stages:First,the grayscale,the gradient map and the saliency map of the distorted image are calculated respectively.The saliency map is obtained by using scale-invariant feature transform(SIFT),then the grayscale image,the gradient map,and the saliency map are divided into blocks of the same size;Second,the blocks of the gradient map are converted into discrete cosine transform(DCT)domain,and the RFSV is designed in DCT domain.The sum of the RFSV are used to measure the blur degree of image;Third,in order to reduce the impact of the image content,the variance of grayscale image and the DCT domain entropy of gradient map are used to normalize the RFSV;Fourth,the number of SIFT feature points in the saliency map is used to assign an independent weight to each image block.The method can evaluate images of different blur types and has good robustness.(2)Aiming at the blur in video,a video deblurring method based on motion vector(MV)is proposed.Motion vector is used to search image blocks quickly,and the improved convolution neural network is used to achieve video deblurring.The proposed method consists of four stages:First,the proposed blur image quality assessment algorithm based on response function of singular values is used to locate the blurred image blocks in video frames;Second,motion vector is used to search the candidate blocks corresponding to the blurred image blocks,and the objective function is used to obtain the optimal candidate blocks;Third,The blurred image block and the optimal candidate block are taken as sample inputs to convolution neural network,then the restoration result of the blurred image block is obtained;Fourth,the blurred image block is replaced with the restoration result,and the boundary artifact is removed to obtain the restoration result of the whole video frame.Experiments show that this method can get clear restoration results and its overall performance is better than other methods.