Reconstruction Algorithms Of Industrial CT With Truncated Projection Data

Computed tomography (CT) is extensively applied in industrial, medical and safety inspection areas for its advantages of nondestructive, high precision, visualization. CT image reconstruction is a process of getting the slice image from its projection data. When the projection data is complete, it is easy to get high quality reconstructed image and the reconstruction algorithms have been studied extensively. But when the projection data is incomplete, the truncated problems such as interior problems, exterior problems and limited angle problems will arise. CT image reconstruction with truncated projection data is still one of the research hotspots.The same material in a casting (or the same tissue in human body) has identical or similar attenuation coefficient. That is to say, CT image is of piecewise-constant values approximately and its gradient image is sparse. In recent years, based on this assumption, regularization algorithms based on total variation minimization (TVM) have achieved approval results for few-views problems. They are also effective for limited angle problems, except that there are still gradual changed artifacts in the nearby regions of strong edge. In order to solve the problem, we investigate a slide correction method, which correct the artifacts through a process like"slide". Introducing the slide correction method to TVM regularization algorithms, we get a new algorithm, which is referred to as slide corrected-TVM (SC-TVM). Experiments with simulated and practical data show that SC-TVM can reduce the artifacts and improve the quality of reconstructed images.Pipe-like objects (such as pipeline) are widely used. The nondestructive testing of them has significant meaning to the reduction of accidents and economic loss. When the projection data is complete, X-ray CT can exactly reconstruct defects in the objects. But when the chords of pipe wall are too long and the energy of X-rays is not adequate enough, the X-rays will be attenuated completely and the projection data for this part will be missing. This is the truncation problem caused by the long chords of pipe wall. Exterior problem of pipe-like object will happen when the object is so large that x-rays cannot cover half of the whole slice and then offset the detectors. To some extent, the above two problems both belong to limited angle problems because of the fact that every points in interested region can only be covered by the X-rays with limited angle. This truncation can blur the edge of CT image and lead to artifacts. Combined with the prior information of pipe-like objects, such as there interior and outer diameters, TVM algorithms can get reconstructed images with artifacts. In order to solve the problem, we investigate a subregion averaged correction method, which divides the preliminary image into some subregions using Chan-Vese (C-V) active contours model, and then averages the values of all points in each subregion respectively. Introducing the subregion averaged correction method to TVM regularization algorithm, we get a new algorithm, which is referred to as subregion averaged-TVM (SA-TVM). Experiments with simulated and practical data verify that SA-TVM can improve the quality of reconstructed images.In linear scan cone-beam CT, the locus of x-ray source and detector array relative to object is a straight line. It is simple and easy to realize. However, its projection data is always incomplete because of the limit of the cone-beam angle along the direction of motion. In this situation, the common analytic reconstruction methods cannot get high quality images. Helical scan cone-beam CT can get high quality volume data with high longitudinal resolution and is suitable for the long objects. But, when the cone-beam is too narrow to cover the cross section of the object, the projection will be truncated. Then the reconstructed results of common analytic reconstruction methods will introduce artifacts. Iterative reconstruction algorithms have advantages of noise suppressing, artifacts reduction, easy to handle projection data truncation, and can introduce the prior information of objects. But they are time consuming, which is an important bottleneck of their applications. Parallel computing is a good approach to manage this problem. We applied simultaneous algebraic reconstruction technique (SART) to linear and helical scan cone-beam CT, and focus on the parallelization of SART. In the parallelized SART of linear scan cone-beam CT, we investigate a balanced loads trapeziform decomposition scheme and subsets-reduce communication technology. In the parallelized SART of helical scan cone-beam CT, we investigate an image space decomposition scheme, subsets-reduce communication technology of projection data in a view and communication technology of image data between two adjacent views. We also give theoretical analysises of their parallel performances, respectively. The implications of parallel SART are realized in a cluster and the results verify their correctness and validities.