Space-time super-resolution from multiple-videos

In this paper, a new method for reconstructing a video of higher spatial and temporal resolutions from multiple low-resolution video sequences is introduced. The proposed cost function includes a new space-time regularization term based upon Huber-Markov random field (HMRF) model which is convex but non-quadratic. However, this prior can be efficiently and accurately approximated by a quadratic form through an iterative process. This regularization form is a type of variational integral that exploits the piecewise smoothness nature of high-resolution images. We also address in detail the main reason for appearance of the so called “ghosting effect” in the temporal super-resolution reconstruction and explain how it can be resolved.