| Computed Tomography(CT)imaging is a technique for obtaining a cross section of an object to be inspected by X-ray scanning,which can display the internal structural information of the object intuitively and accurately.However,in the practical application of CT imaging,it is often faced with the challenge of under-sampling and data incompleteness:the atmosphere of industrial CT is very complicated,and the completeness of projection can hardly be guaranteed due to scanning conditions and the shape,size of the detected object;in medical CT,how to use less projection angles to minimize the radiation dose of X-rays,and reconstruct higher quality images has always been a hot issue in the field of CT research.By related literature survey and communicating with CT engineers and technicians,this paper defines the projection with 20 or less projection angles in[0,360°)(i.e.the projection angle interval is not less than 18 degrees)as ultra-sparse projection.CT image reconstruction algorithms can be divided into two categories:analytical method and iterative method.Analytical method is generally applied to the image reconstruction of complete sampling data.Reducing the number of projection angles will lead to serious stripe artifacts in the reconstructed image,and the image quality will be greatly reduced.Iterative method transforms the CT image reconstruction problem into a multi-dimensional under-determined equation problem,and realizes image reconstruction by searching for the combination of pixel values satisfying certain convergence condition through continuous iteration.Compared with the analytical method,iterative method has a natural advantage in the reconstruction of sparse view sampled data.In particular,with the emergence of the compressed sensing theory,some sparse prior information inherent in the image is integrated into the iterative reconstruction algorithm,which lays a theoretical foundation for the use of ultra-sparse view projection to reconstruct high quality images.In this paper,experiments are carried out on the CT image reconstruction of ultra-sparse projection.Firstly,two-dimensional image reconstruction of fan-beam scanning data is studied.We combine iterative algorithm with compressed sensin~g theory,and use the total variation of image as objective function of convex optimization.A two-dimensional CT image reconstruction algorithm for ultra-sparse projection is proposed,called USP-2DTV in this paper.Then we extended the USP-2DTV algorithm to three-dimensional CT image reconstruction,and the USP-3DTV algorithm is proposed.The experimental results show that under the condition of ultra-sparse projection,the proposed algorithms can obtain high-quality reconstruction images.which is of some research potential and practical value.In addition,the USP-2DTV algorithm in this paper can process a minimum of 18 projection angles,and the required scan time and radiation dose are 50 times less than that of traditional FBP algorithm(which usually uses 1000 projection angles).In order to improve the efficiency of the iterative algorithm,based on the CUDA architecture of NVIDIA,this paper realized the CPU+GPU heterogeneous parallel computing of the USP-2DTV algorithm,which improves the efficiency of the image reconstruction program to 2.5 times that of the CPU serial computing.This thesis combines CT iterative image reconstruction technology with compressed sensing theory to study the 2D/3D CT image reconstruction of ultra-sparse projection,and realizes the CPU+GPU heterogeneous parallel computing of the algorithm,which has certain academic exploration significance and engineering application value. |
PDF Full Text Request