A novel non-local means algorithm based on rotation and mirroring transformation for sparse-view CT reconstruction

To solve the problem of sparse-view CT reconstruction, an adaptive NLM reconstruction algorithm based on rotation and mirroring transformation (RAMNLM) is proposed. In RAMNLM, a novel similarity measure based on rotation and mirroring transformation is designed to avoid over-smoothness. Moreover, a new filter parameter in RAMNLM, which benefit preserving of structural information would change adaptively with the iteration number and vary with the pixel gradient. The proposed CT image reconstruction mainly consists of three steps: Algebraic reconstruction technique (ART); Positivity constraint; RAMNLM filtering. The above steps alternated each other until the convergence criterion is satisfied. The proposed algorithm is validated on Shepp-Logan phantom. The simulated reconstruction result demonstrates that the ART-RAMNLM method could suppress noises more effectively, restore image details better, and improve the image quality significantly. (author)