Graphics processor programming to speed up iterative image processing algorithms

Abstract

Backprojection is an algorithm for calculating an estimation of an object given the data of its computer tomographic scan. The standard algorithm currently applied by most programs is only using the CPU as computational object, what takes tenth of minutes, which is fairly slow for industrial standards. This paper is concerned with the improvement of the backprojection algorithm by exploiting the highly parallelizable nature of the algorithm and increasing the computation time by adding the internal GPU as computational unit.