In this paper, a joint-domain dictionary learning-based error concealment approach is proposed. We extend the existing joint-domain dictionary learning methods to make more suitable scheme for error concealment. The main idea is to train an offline over-complete dictionary pair and learn two mapping matrices using two sets from the original and corrupted patches in a coupled manner, such that the sparse representations of corresponding patches are the same in a common domain. At the concealment process, the dictionary corresponding to the corrupted data set is used to compute the sparse representation of the corrupted patch, which is further transformed using the respective mapping matrices in order to find a good estimation of the original patch. Then, the dictionary corresponding to the original data set is used to recover the patch. In addition, we propose one modification to the recovery process by benefiting of non-local similarities in the images. Comparisons with several state-of-the-art approaches demonstrate the efficiency of the proposed approach for image error concealment, both in terms of quality and complexity.