Blind Deblurring From Single Motion Blurred Image

Image is an important source for human to access to information and which occupies an important position in modern society. However, the acquired images are always blurring duo to camera shake or the movement of objects in the scene during the capture process, and which seriously affects the use of the image and its subsequent processing. The purpose of image de-blurring is to recovery clear images that contain a wealth of information from blurred ones by a certain method. When the blur kernel is unknown, image deblurring problem becomes to be the so-called blind image deblurring problem, and which is a very challenging ill-posed problem. Image deblurring is an important topic in digital processing and computer vision communities and which is widely used in areas such as photography, optics, astronomy, medical images, monitoring, remote sensing and military research et al. It’s also a hot topic of academia in recent years. Therefore, it has important theoretical and practical significance.This paper focuses on the single image based blind deblurring problem. First, we introduces the research background and significance of image deblurring problem, and a detailed analysis of the development is followed. Second, we introduce the degradation model that leads to blur, which includes the continuous model and the discrete model. Followed by the introducing of the probability model for image deblurring, several noise models and priors. Several common blind deblurring methods are also been analyzed. Then, we propose a novel robust blind de-blurring method from single blurred image. We propose to use10-constraint smoothing to select salient edges which further increase the robustness of kernel estimation. An effective normal-ized sparsity measure is employed to constrain the latent image restoration, which can preserve more natural image properties, e.g., sharp edges, tiny textures. Furthermore, we also provide an effcient algorithm to solve the proposed models.Experiments show that our method is effective for most of uniform motion blurs and the quality of blur kernel and latent image is better than traditional methods, and the speed of the algorithm is also been improved significantly.