UHD Video Super-Resolution Using Low-Rank and Sparse Decomposition

Sparse coding-based algorithms have been successfully applied to the single-image super resolution problem. Conventional multi-image super-resolution (SR) algorithms incorporate auxiliary frames into the model by a registration process using subpixel block matching algorithms that are computationally expensive. This becomes increasingly important as super-resolving UHD video content with existing sparse-based SR approaches become less efficient. In order to fully utilize the spatio-temporal information, we propose a novel multi-frame video SR approach that is aided by a low-rank plus sparse decomposition of the video sequence. We introduce a group of pictures structure where we seek a rank-1 low-rank part that recovers the shared spatiotemporal information among the frames in the group of pictures (GOP). Then we super-resolve the low-rank frame and sparse frames separately. This assumption results in significant time reductions, as well as surpassing state-of-the-art performance both qualitatively and quantitatively.