Fast Blind Motion Deblurring With Split Bregman Method

Image deblurring is one important field of image processing. After years’ researching on this, more and more methods of image restoration have been proposed. While the practicability for some of them haven’t been proved and are also to be improved, image deblurring has a critical research values both on its theoretic and practical.In this dissertation, after studying the fast blind image restoration algorithms, kernel estimation technology and split Bregman methods, a new algorithm based on split method is proposed to solve the blind motion deblurring problem, which uses multi-scale strategy and image edge to estimate the blur kernel and establishes a fast blind motion deblurring frame. Experiment results show that our method have a superior effect both in computation time and visual effect. The main work is divided into five parts as follows:(1) In introduction, the meaning and purpose of image deblurring is talked about, the overview and development of single motion blurred image restoration technology is analyzed and the framework of our dissertation is given.(2) Image degradation model and the relative basic knowledge as well as the traditional methods of image restoration is studied, then based on this all the existing algorithms are classified into two main parts and comparison is made in this part.(3) Fast motion deblurring method is studied, which includes blur kernel estimation and image restoration. When estimating the blur kernel, first bilateral filter is used for image denoising, then improved shock filter is for edge enhancement, useful edges information is selected to have a blur kernel, after adjustment, a high-quality blur kernel is obtained. In image restoration, fast motion deconvolution is used to get a clean image.(4) Moreover, based on researching the principle and property of split Bregman iterative algorithm, a new fast motion deblurring method is proposed which combines fast motion deblurring and split Bregman iterative algorithm as well as the previous blur kernel to get a clean image. It mainly focuses on the restoring the details of image and increasing the speed of iteration. Experiment results show the effectiveness and superiority of our algorithm both in computation time and visually.(5) Last is the summarization and outlook of image restoration.