Sparse Angle CT Image Reconstruction Based on Mixed Regular Terms
To solve the problem that the traditional iterative algorithms can not guarantee the reconstruction of high-quality medical CT images in the case of data undersampling,a method of double regular term penalty least squares based on surface area minimization and Group Sparse Representation(GSR)was proposed for sparse angle CT image reconstruction.According to the rule that the surface area of the noisy image is larger than that of the smooth image in the grid field,the surface area term was used as a prior constraint on reconstructing the image to achieve the purpose of smoothing the image noise;based on the feature that GSR preserves the fine structure of the image by using non-local similarity,the reconstruction model was established by using GSR as another prior constraint.An alternate minimization scheme was used to decouple the reconstruction problem into solving the intermediate image subproblem,and solving the dictionary and sparse representation subproblem.Then,the two sub-problems were further solved by the half-quadratic splitting method and the split Bregman method respectively.The experimental results show that compared with the algorithms PLS-GSR,PLS-NLM and SART-GSR,the reconstructed image of the proposed algorithm has good visual effect,which can clearly identify the internal structure and details of the image,and the value of FSIM,PSNR and MSE are superior to the comparison algorithm.
group sparse representationsurface areaimage smoothingnon-local similarityimage reconstruction
万方数据
维普
