An anisotropic alternating regularization-based reconstruction algorithm for cone beam computed laminography

Abstract

Computed laminography (CL) is a promising tool for non-destructive visualization of the internal structures of relatively flat objects. However, CL scanning characteristics lead to the absence of projection data along certain directions. When applying conventional algorithms to cone-beam CL projections, the reconstructed images will be deteriorated by cone angle artifacts. Motivated by this finding, we proposed a CL reconstruction model based on anisotropic alternating regularization (AAR) and its numerical solver. The model contains three regularization terms. l0 norm of image gradient is employed as the x- and the y-direction regularizers to perform edge-preserving smoothing. And relative total variance (RTV) is used as the z-direction regularizer to recover inter-slice structures. Experiments on both disk and PCB phantoms demonstrate that our algorithm is effective at preserving image edges, eliminating inter-slice aliasing and denoising.