CT Reconstruction Algorithm Based On Novel Image Non-local Self-similar Low-rank Model

Computed tomography(CT)is one of the most important and widely used aids in the diagnosis of diseases in the medical field.Existing CT imaging requires a long period of X-ray scanning on the human body,which will increase the risk of human cancer with a high probability,but short-time scanning will lead to poor CT imaging quality.Improving the quality of CT images from the perspective of improving hardware equipment will always bring unaffordable costs to people.Therefore,from the perspective of algorithms,reconstructing low-quality CT images to obtain high-quality images is the current mainstream improvement direction.Existing image reconstruction algorithms often only consider the use of image sparsity prior features and ignore non-local self-similar features of images,resulting in low imaging quality.With the development of image reconstruction technology,the emergence of many low-rank models also brings new thinking directions for image reconstruction algorithms.Aiming at the problem that traditional CT image sparse representation models often only consider the sparse characteristics of the image,usually assume that the sparsely coded image blocks are independent,and essentially do not consider the correlation between similar blocks.A model for transforming the sparse coding problem into a low-rank minimization problem.Based on the application of image sparse characteristics,this model fully considers the non-local self-similar characteristics of images to construct an image group matrix.Due to the strong correlation between the column vectors obtained from the expansion of each similar image block,the constructed image group matrix has a "low" The sparse coding problem of original CT image reconstruction is transformed into a low-rank minimization problem,which is more convenient to apply a new low-rank model to solve the problem.In order to solve the problem of low-rank minimization,the traditional kernel norm minimization method considers all singular values of the matrix equally,resulting in the emergence of biased solutions.In this thesis,from the perspective of using a non-convex rank relaxation function,a new low-rank model schatten p is applied.-Kernel norm,a CT image reconstruction method based on weighted schatten p-kernel norm regularization is proposed to relax the image reconstruction optimization problem.Then,from the perspective of directly improving the traditional nuclear norm,the truncated nuclear norm is applied and combined with the model constructed in this thesis,a CT image reconstruction method based on the regularization of the weighted truncated nuclear norm is developed.The structural information of CT images is well preserved.Finally,by comparing with SART(Simultaneous Algebraic Reconstruction Technique),GSR-SART(Group-Sparsity Regularization-SART)and traditional NNM(Nuclear Norm Minimization)methods,simulation experiments show that the PSNR value obtained by the two algorithms proposed in this thesis is at least improved.1d B,the SSIM value is closer to 1,and the reconstructed image residual is smaller,which can better preserve the image details and edge structure information.