Sparse coding based on L2,1-norm and manifold regularization for remote sensing images

Sparse coding is an important tool in image re- construction. In this paper, a novel sparse coding algorithm based on L2,1-norm and manifold regularization is developed to reconstruct remote sensing images. In the proposed algorithm, loss function based on L2,1-norm has the capability of reducing outliers. Dictionary and sparse coding based on L2,1-norm can select features with joint sparsity. Manifold regularization can retain more geometric information. An efficient algorithm for updating dictionary is developed, and the convergence is proved. To this end, we compare the proposed algorithm with other existing methods. Experiments on real remote sensing images demonstrate that the proposed algorithm has promising reconstruction results.