Super-resolution image reconstruction via adaptive sparse representation and joint dictionary training

Recently, sparse representation has emerged as a powerful technique for solving various image restoration applications. In this paper, we investigate the application of sparse representation on single-image super-resolution problems. Considering that the quality of the super-resolved images largely depends on whether the employed sparse domain can represent well the target image, we propose to seek a sparse representation adaptively for each patch of the low-resolution image, and then use the coefficients in the low-resolution domain to reconstruct the high-resolution counterpart. By jointly training the low- and high-resolution dictionaries and choosing the best set of bases to characterize the local patch, we can tighten the similarity between the low-resolution and high-resolution local patches. Experimental results on single-image super-resolution demonstrate the effectiveness of the proposed method.