Magnetic resonance image reconstruction via the augmented TV-regularized problem

In order to perfectly recover MR image using compressed sensing (CS) with TV-regularization term, one has to minimize ill-conditioned optimization functions on large datasets, which are usually not tractable. To address this problem, we propose a novel algorithm for solving an augmented total variation (TV) regularized model with linear constraints, which can obtain accurate solutions. The classical alternating direction method of multiplies (ADMM) framework is utilized to reformulate this model. A linearization strategy is designed to simplify operation. In addition, we combine weighted solution procedure as well as backtracking strategy for improving the reconstruction quality. Experimental results demonstrate that the proposed approach achieves better reconstruction performance than traditional CS-based methods.