high performance computing on graphics processing units: hgpu.org

Iterative reconstruction algorithms augmented with regularization can produce high-quality reconstructions from few views and even in the presence of significant noise. In this paper we focus on the particularities associated with the GPU acceleration of these. First, we introduce the idea of using exhaustive benchmark tests to determine the optimal settings of various parameters in iterative algorithm, here OS-SIRT, which proofs decisive for obtaining optimal GPU performance. Then we introduce bilateral filtering as a viable and cost-effective means for regularization, and we show that GPU-acceleration reduces its overhead to very moderate levels.