Pansharpening via Locality-Constrained Sparse Representation

Recently, sparse representation based approaches have been shown an effective performance for pansharpening. However, these methods imposed `0 or `1 -norm constraints on the sparse coefficients. The local similarity of sparse coefficients was ignored. Motivated by the importance of data locality, in this paper, we propose a locality-constrained sparse representation algorithm for pansharpening, which keeps the data locality during the sparse representation process. The learned dictionary is able to preserve local data structure, resulting in improved data representation. During the sparse coding stage, analytical solutions are provided based on the basis of mathematic deduction. The pansharpening results show that the proposed method is competitive to the other well-known fusion methods.