In this letter, a method based on block-matching and Low-rank matrix recovery (BM-LRMR) is proposed for single image reconstruction. While existing approaches usually use many image sequences to restructure new images, our method tries to use a single image to realize the image reconstruction. In our algorithm, the single image is firstly partitioned into blocks. Then a block-matching technique is employed in grouping and constructing the approximately low-rank matrix. A low-rank matrix recovery is used in low-rank matrix decomposition. Finally, a weighted average strategy is using to compute a final estimate of the image by aggregating all of the obtained local estimates. Experimental results on different applications demonstrate the proposed model can work excellently for single image reconstruction.
Nonlocal latent low rank sparse representation for single image super resolution via self-similarity learning