Tomographic Reconstruction Via 3D Convolutional Dictionary Learning

摘要

Convolutional sparse representation is an efficient tool regularly applied to image processing problems. In applications where the signal domain and measurement domain differ, such as computational tomography, additional effort is needed to apply the convolutional sparse representation model. We develop a novel convolutional dictionary learning framework that constructs a sparse representation in the signal domain and allows for accurate signal reconstruction from incomplete sets of measurements. Application of the proposed method on simulated parallel beam tomography with synthetic and experimental data demonstrates performance comparable to that of state-of-the-art reconstruction techniques.