3D Image Reconstruction from Limited Projections by Simulated Annealing

CT (computed tomography) is a powerful technique for obtaining a volume image in medicine. It has also been applied to many other research fields, such as defect detection in industry and nuclear physics. Since in many applications, the viewing directions are limited, we have to reconstruct a image from a few limited number of projections. In this paper, we propose a new robust CT image reconstruction method by simulated annealing. The volume image is represented by a set of 3D Gaussian functions and the reconstruction is done by estimating parameters of kernel functions (3D Gaussian functions). A fast algorithm is also proposed to reduce the large computation time.