Image sparse representation with local ARMA and nonlocal self-similarity regularizations for super-resolution

摘要

Since the single image super-resolution (SR) is an extremely ill posed problem, we introduce a novel auto regressive moving average (ARMA) model-based regularization term into the spare representation-based framework to deal with it in this paper. In our framework, we have a dual regularization. Firstly, we use the ARMA models trained from external samples to establish a regularization term. ARMA model-based regularization serves as a local constraint. Secondly, we introduce the nonlocal (NL) self-similarity as another regularization term. Both the local and the NL regularizations are unified into the sparse representation-based framework. Finally, extensive experiments verify the effectiveness of the proposed method. (C) 2017 Elsevier B.V. All rights reserved.