Real-time super-resolution over raw video sequences

Abstract

This paper addresses the enhancement of the spatial resolution of a video sequence from a low resolution video sequence in real time. The technique used, known as super-resolution reconstruction, exploits the relative motion from frame to frame that produces sub-pixel shifts. The algorithm, based on a previous version mapped onto a video encoder architecture, is oriented towards a hardware implementation and requires resources optimization. In order to achieve a good resolution improvement, the motion estimation algorithm must produce motion vectors as close to the real ones as possible. At the same time, this motion estimation must match real time requirements. Therefore, an exhaustive technique is applied in combination with a simple segmentation of each frame for a motion prediction refinement. Experimental results have been obtained for a set of video sequences subjected to different motion characteristics.