Blind Image Debluring Via Adaptive Sparse Image Modeling

Image blur is a common image degradation problem, and restoring the original image from a blurred image is quite important due to many photo scenario is non-reproducibility. Traditional methods cannot cater the clear image because the energy of their regular term can get the minimum value when the solution is the blurred image. In order to solve for the clear image, many algorithms add some complex process to sharpen the image. As the result, the deblurring image would seem satisfactory, but the solving process is too complicated and it’s difficult to ensure its feasibility in the theory.In order to overcome the shortcomings of the traditional deblurring methods, we must analyze the priori information of the natural images, and estimate effective regular constraints for deblurring due to natural regularization plays a key role in image restoration. The main works and the contribution of the thesis are as follows:1. In order to overcome the shortcoming that traditional methods cannot obtain the clear image in the process of deblurring, the thesis proposed Edge-adaptive sparse representative based image deblurring algorithm. The algorithm has the following characteristics: 1) The algorithm constructed the regular term in image gradient domain, and we removed the average image from the image gradient value, which improves the image edges and the details clearly. 2) This algorithm uses adaptive weighted coefficient to punish the gradient regularization, so the energy of the regular term in the algorithm model can be minimized when the solution is the clear image. At the same time, the algorithm model can be interpreted with the Bayesian Theory, which illustrates the effectiveness of the proposed model in theory. The experimental results also prove the effectiveness of the algorithm.2. In order to improve the performance of the deblurring method based on edge adaptive, the thesis proposed Self-adaptive sparse representation based image deblurring algorithm. The algorithm adds a sparse regularization of the image standard deviation, which not only keeps the minimum regular term energy favoring the clear image, butalso makes the image parameters calculated with a closed form. So we can update the parameters more precisely. In the whole process of the optimization, the parameters calculated with image local information that doesn’t depends on the initial solution of the clear image, so we can solve the problem faster. Finally, we test our algorithm on synthetic blurring images and real-world blurring images. The results prove that our algorithm is quite good for deblurring image.