Multi-orientation depthwise extraction for stereo image super-resolution

Xiangyang Fan, Renjia Ye, Feifan Cai, Jinhua Liu, Yuhang Li, Liting Huang, Youdong Ding

Abstract: The increasing trend of binocular imaging in recent years has sparked a surge of interest in stereo image super-resolution. While considerable progress has been made in improving model performance, the potential of single-view and cross-view features remains largely unexplored. To address this, we present a novel network that incorporates both intra-view and inter-view feature extraction to enhance stereo image super-resolution. Specially, we design a multi-orientation depthwise extraction module to sufficiently extract various orientation features within a single view. Additionally, a cross focus module is proposed to capture more reliable hierarchical cross-view features. These modules can be integrated together to exploit trustworthier complementary information for HR image reconstruction. Our experimental results showcase the excellent performance of our method, surpassing all previous state-of-the-art methods for stereo image super-resolution.

External IDs:dblp:journals/sivp/FanYCLLHD23