The Study Of TCT Statistical Reconstruction Algorithm Based On SPS

Reconstruction algorithm plays a very important role in transmission computedtomography. When the projection data is incomplete or has serious noise, the quality ofreconstructed image by statistical iterative reconstruction algorithm, in which constraint andprior information can be joined, is better than the traditional analytic reconstruction algorithm.However, the speed of statistical iterative reconstruction algorithm is slower than analyticreconstruction algorithm. To the question of speed and quality of reconstruction, this paperconsiders both sides of algorithm accelerating and Maximum-A-Posteriori estimate, and somemain studies are as following:Firstly, this paper introduces TCT statistical reconstruction algorithms based on themaximum likelihood estimate: convex algorithm, SPS algorithm, and through theoreticalanalysis and experimental comparison, the smaller error and bigger correlation ofreconstructed image by SPS algorithm is than convex algorithm in the same iteration time.Secondly, to the shortcoming of slow convergence speed for ML-SPS algorithm, thispaper considers algorithm accelerated, and ordered subset is applied to the algorithm andextended to three-dimensional cone track. However, because OS-SPS algorithm does notconvergence, this paper presents an improved relaxation factor sequence. The experimentalresults show that this method does not only ensure convergence, but also suppress the spreadof noise.Lastly, to the question of projections containing noise, this paper studies OS-SPSstatistical reconstruction algorithms based on the Maximum-A-Posteriori estimate. On the basisof“transform domain”and statistic characteristics of wavelet coefficients and energy balanceprinciple of different scales, high frequency information is reasonably processed in the processof iterative reconstruction. Meanwhile, the second prior function and wavelet prior function areapplied to MAP-OS-SPS reconstruction algorithm from the airspace and frequency domain.The experimental results show that wavelet prior does not only suppress the spread of noise,but also save the edge information of reconstructed image, and the quality of reconstructedimage is better than second prior.