Blind Image Deblur Method Based On Image Edges And Low Rank Fusion

Recovering an unblurred image from a single motion-blurred picture has long been a fundamental research problem. Image deblurring can be further categorized to the blind and non-blind cases. Main Problems in image deblurring are reducing the ringing effects which may occur in the strong edge area, suppressing noise and saving computational cost. Because the blur kernel and the clear image in bind image deblur method are unknown, the problem becomes very challenging. In this paper, we focus on the research of blind image deblur method, we propose three blind image deblur methods based on the edge information of the image and low rank fusion.First: In our work, we explore a novel blind image deblur method based on normalized sparse regularization. Compared with previous methods, the proposed method introduces the standard sparse regularization into the regularization of the image block. The proposed method can suppress the ill-posed problem. The strategy tailored towards the estimated blur kernel to a real blur kernel in the process of coarse to fine fashion. The experiment results show that the proposed method performs better than some state-of-the-art results. There are little artifacts in clear images.Second: Based on the first work, we propose a novel image blind deconvolution method based on edge information. This method exploits the characteristic of Wiener filtering that not only preserves the image edge details of the high frequency part of the image, but also carrys out the characteristics of the low frequency part of the image, and eliminates the impact of the narrow edge. In details, wiener filtering is added to the current popular iterative framework of Pyramid. After each iteration, a clear image is generated, and then the image is processed by Wiener filtering. We get the new clear image as the initial image of the next iteration, and then get a more accurate kernel. Finally, with a non-blind deblurring algorithm, a faithful latent image can be obtained.Third: In our work, we explore a novel blind image deblur method based on low rank fusion. The proposed method attempts to estimate multiple kernels using any blind image deblur method. Those kernel have some similarity in structure, because these kernels are estimated by the same blind image deblur method. However, there is still a little difference in details, which can be used to make a more robust kernel with low rank fusion. Meanwhile, noises and singular values existed in the process of solving blur kernel can be reduced. The experiment results demonstrate that the proposed algorithm can reduce the inaccuracy of the kernel estimation.