A comparative study of popular interpolation and integration methods for use in computed tomography

We compare various popular methods available for projection and backprojection in CT. Assuming linear rays and a simple density integration along them, we consider both line- and area-based methods. Here, two key components govern the quality of a projection result, given the discrete nature of the data and reconstruction result: interpolation and integration. Both of these are studied here. In order to separate these fundamental issues from those related to perspective fan and cone-beam effects, we restrict ourselves to a parallel-beam projection geometry. We also compare these different methods in light of a possible efficient implementation on programmable commodity graphics hardware (GPUs). To this end, we propose a new method for interpolation based on hexagonal sub-sampling, which achieves superior results. In order to achieve a data-independent comparison, we employ a dataset of very high and uniform frequency content, the so-called Marschner-Lobb dataset.