Research On Hybrid Constraint Based Blind Motion Image Deblurring

Motion blur is one of the most common blur types.Camera shake and the relative motion of the object and the camera are two main reasons causing the motion blur.Blind image deblurring is a process of deblurring while the point spread function is unknown.It is a severe ill-posed problem,which can be effectively transformed to a well-posed problem by using regularization method.By deeply studied the methods of kernel and image regular term establishing,we proposed a hybrid constraint based method for blind motion image deblurring,based on the classical regularization method.From the aspect of model,in order to keep the sparsity and continuous of the motion blur kernel,we used the hybrid constraint to constrain the blur kernel,besides that,a nonconvex higher-order total variation was introduced to constrain the image in the model,which can eliminate the staircase effect and protect the image edge.From the aspect of the algorithm,we proposed iterative reweighted minimization based and split Bregman iteration based method to solve model.The main research works carried out in this thesis are as follows:(1)In this thesis,we made a survey on blind image deblurring technology development status,especially the regularization based method.We summarized the basic theory of blind image deblurring,the characteristic of the motion blur kernel and the difficulties of the blind motion image deblurring.(2)A deeply study on establishing blur kernel regular term and image regular term was made in this thesis.We summarized the common methods of blur kernel establishing and anlasyzed the advantages of H1 norm and L1 norm hybrid constraint and its ability for keeping the sparsity and continuous of the motion blur kernel.Summarized the common methods of image regular term establishing and anlasiyzed the adavantages of nonconvex higher-order total variation and its ability for eliminating the staircase as well as protecting the image edge.(3)We proposed the hybrid constraint based blind motion image deblurring model and the corresponding algorithm in this thesis.In the model,in order to keep the sparsity and continuous of the motion blur kernel,we introduced the H1 norm and L1 norm hybrid constraint on blur kernel.In order to eliminate the staircase effect as well as protect the image edge,we introduced the nonconvex higher-order total variation constraint on image.To solve the model,we proposed the iterative reweighted minimization based and the split Bregman iteration based algorithm.In the algorithm,there are two iterative process.In the outer iterative,we used the iterative reweighted minimization algorithm to update the weight.In the inner,the split Bregman iteration based algorithm was used to caculate the blur kernel and the latent clear image.Beside that,in order to improve the quality of the algorithm,we added the nonnegative constraints on the image,normalized and dynamic threshold constraints on the blur kernel in the algorithm.(4)At the end of the thesis,the proposed method was tested on both synthetic and real-world blur images.The results was compared with some classical blind image deblurring method and demonstrated that the proposed method outperforms the previous representative methods in both quality and time costing.