Non-Blind Image Deblurring Method Based On Shear Total Variation

Images are the most intuitive way for people to obtain information about the outside world,but since different levels of blurred noise are generated due to different conditions of contamination during image generation,storage and transmission,how to remove different types of blur kernel is an important and meaningful research task in image processing.Image deblurring is the process of recovering as much of the original image as possible from the degraded image.In the past,the classical total variation and associated deformation regularization models proposed by Rudin-Osher-Fatemi were applied to the image deblurring problem.However,due to the limited selection of a priori information of the original image and the inevitable existence of stair-step artifacts in the image recovery process,many scholars have done different researches to improve the model.I was inspired by the previous research and proposed a relevant model for solving the problem by introducing the shear operator and combining it with the first-order total variation and high-order total variation models using the alternating direction multiplier method(ADMM)and the fast solving property of the BCCB matrix,respectively.First,this thesis introduces the background of image deblurring and its significance,the current status of domestic and foreign research and related theoretical knowledge.Secondly,this paper gives an overview of the first-order total variation,higherorder total variation,shear operator and BCCB matrix.In this paper,the first-order total variation regularization model of the total variation operator and the higher-order total variation regularization model of the shear operator are proposed considering the advantages of the total variation and the shear operator,and the model is solved by applying the properties of the BCCB matrix and the ADMM method.Experimental results show that the proposed model can not only recover relatively clear images well,but also better recover image edges and reduce staircase artifacts,and our proposed algorithm achieves satisfactory results by comparing the visual effects and quantitative measures with other advanced methods.