Cross View Capture for Stereo Image Super-Resolution

Stereo image super-resolution exploits additional features from cross view pairs for high resolution (HR) reconstruction. Recently, several new methods have been proposed to investigate along epipolar lines enhance the visual perception of recovered HR images. Despite impressive performance these methods, global contextual images are left unexplored. In this paper, we propose a capture network (CVCnet) stereo by using both and local extracted views. Specifically, design block diverse feature embeddings views in vision. addition, cascaded spatial module is redistribute each location maps according weight it occupies make extraction more effective. Extensive experiments demonstrate that our CVCnet outperforms state-of-the-art achieve best tasks. The source code available at https://github.com/xyzhu1/CVCnet.