Research On Computer Tomography Image Reconstruction Algorithm Based On Compressed Sensing

Reconstruction algorithm is the most essential part of CT system,which is meaningful to research.The traditional algebra reconstruction algorithm(ART)can not effectively reduce the noise due to the incomplete data and the radiation is very harmful to patients through CT scan.Thus,based on the compressed sensing which is able to use undersampling to recover the high quality image,research about computer tomography image reconstruction algorithm based on compressed sensing begins.First of all,principle of CT imaging,basic CT reconstruction algorithm and theory of compressed sensing are talked about.Based on that,research about projections onto convex sets(POCS)algorithm has been done,discussing about iterations and lambda whether they can influence the reconstructed image.Next,this thesis finished the algorithm based on total variation(TV)and projections onto convex sets because total variation can keep the edge information of things.Experiment analyses whether sigma and iterations will affect the results.In order to get convergence easily in POCS-TVM,a method based on Armijo rules is then researched,which gets the adaptive steps automatically.And a compressed sensing algorithm based on the prior image(PICCS)is presented and its experiments show that prior images have different influence on constructed images.Finally,after analysing the influence of dictionary sizes and dictionary numbers to the noise,a revised algorithm based on dictionary learning(KSVD)has been proposed.Simulational results indicate that ART can not remove the noise because of undersampling and iterations and lambda do not have obviously effect on it.POCS-TVM can keep the edge of objects.And beta and iterations have the positive effect on improving quality of image,which can remove the noise caused by undersampling and keep the quality of image as well.Armijo-POCS based on Armijo rules is proposed.It is able to calculate the step size automatically and improves the signal noise ratio and speed of algorithm.Then PICCS has been done.If constructed image is similar to the prior image,the result will be accurate but slow,while low quality image will lead to lower quality of constructed images and longer time of algorithm.It is very interesting that result can be acquired with more number of dictionary and the appropriate dictionary size.Based on the experiments,a new method called KSVD-TV based on total variation and smaller dictionary size is presented,which can reduce the noise and keep the feature.It has the higher signal noise ratio than the previous method.