An exponent-based anisotropic variational PDE regularized image reconstruction method for digital tomosynthesis

Abstract

A novel image reconstruction method based on exponent-based anisotropic variational partial differential equation (PDE) for digital tomosynthesis (DT) is proposed. An exponent map instead of fixed values 1 or 2 is used as the order of total variation (TV), which can balance staircase artifact and over smoothing. The smoothed structure tensor (SST) capable of characterizing of different image features is used to build the exponent map, which has good ability of noise suppressing as well as detail preserving. Numerical experiment demonstrates that our method achieves better performance in many aspects (such as noise level, structure similarity) than existing methods, including adaptive steepest descent-projection onto convex sets (ASD-POCS) and selective diffusion regularized simultaneous algebraic reconstruction technique (SART).