Abstract:

Compressed sensing (CS) image reconstruction algorithm is a key point in the CS image acquisition problem. The conventional sparsity based reconstruction algorithms cannot effectively reconstruct various image structures. In order to improve the reconstruction precision with the same number of measurements, the local autoregressive (AR) model and nonlocal selfsimilarity as the additional prior information of image are introduced, and a nonlocal regularized CS image reconstruction model and the relative numerical algorithm are developed. Furthermore, a nonlocal similarity based estimation method for AR parameters is proposed, which are considered in the present reconstruction model. Compared with the traditional sparsity regularized algorithms and related MARX(modelbased adaptive recovery of compressive sensing) algorithm, the proposed algorithm can achieve a higher image reconstruction performance.