A Hybrid System based on Single and Multi-frame Techniques for Super-resolution of Video Sequence

摘要

In one of the methods of multi-frame super-resolution, a high resolution video is reconstructed from a low resolution video of a scene by adding the high frequency components estimated from the sub-pixel shift between the frames using pyramidal tracking and iterative back projection algorithms. In contrast, super resolving a single frame low resolution image is called Example-based super-resolution (EBSR). This is achieved by recovering the missing high frequency contents from a large number of example image patches in a database. In this paper we propose a novel hybrid approach to construct a super resolved video sequence which uses the example-based method for the preprocessing of low resolution video frames as input to the multi-frame super resolution. The approach to use EBSR in the preprocessing stage of low resolution frames acts as an image model error correction process. In this paper, we demonstrate that the proposed hybrid technique yields better results than the chosen stand-alone multi-frame technique.