Image super-resolution via sparse embedding

Although the coupled dictionary pair is jointly trained for image super-resolution, the sparse representations corresponding to the low resolution patches and high resolution patches are inconsistent. Therefore, we propose a sparse embedding method to establish the mapping relationship of the sparse representation. The manifold spaces of the sparse representations between the low resolution patches and high resolution patches are assumed to be similar. The sparse representations of the high resolution image patches can be generated through the optimal weights. The experiment results show that our methods outperform other existing methods.