Multiview Video Super-Resolution via Information Extraction and Merging
Yawei Li, Xiaofeng Li, Zhizhong Fu, Wenli Zhong · 2016
Multiview video super-resolution provides a promising solution to the contradiction between the huge data size of multiview video and the degraded video quality due to mixed-resolution compression. This algorithm consists of two different functional layers. An information extraction layer draws relevant high-frequency information from the high-resolution views via depth-image-based rendering and interpolation. A merging layer fuses multiview high-frequency information to refine the low-resolution view. In this paper, we introduce kernel regression and non-local means to improve the two layers, respectively. Kernel regression adapts to the local image structure and thus outperforms basic interpolation methods. Non-local means exploits the similarity between different views of multiview videos to restore the high-frequency component of a low-resolution image. We constrain non-local means by limiting the pixels used to restore a pixel. The experimental results show the effectiveness of the proposed algorithm.