Scholarlink

<p>Video super-resolution, which has been widely applied in ultra-high definition display, video surveillance, and robot surgery, is the recovery of high-resolution image frame sequences with distinct details and temporal continuity from low-resolution video. Thanks to advancements in deep learning, deep video super-resolution has achieved promising results. However, frame artifacts and shakiness remain key challenges that need to be addressed for further improvements in video super-resolution. In particular, it is still worthwhile to research how to properly extract the spatiotemporal features from successive video frames and efficiently fuse them in order to achieve video super-resolution with rich intra-frame details and consistent inter-frame variations. In light of this, we propose to improve bi-directional recurrent network for video super-resolution with deformable motion alignment structure in this work to precisely estimate the offsets and provide sharper image frames. Experiments demonstrate that the proposed approach outperforms current state-of-the-art methods by 0.51%. The project code is available at <a href="https://github.com/hengliusky/IBRN_VSR">https://github.com/hengliusky/IBRN_VSR</a>.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Improving bi-directional recurrent network for video super-resolution with deformable motion alignment structure

  • 期刊论文

  • 发表日期：

    2025年1月13日

查看全文
  • Heng Liu,
  • Xiangchen Huang,
  • Yuezhong Chu,
  • Mingquan Ye,
  • Tao Liu

摘要

Video super-resolution, which has been widely applied in ultra-high definition display, video surveillance, and robot surgery, is the recovery of high-resolution image frame sequences with distinct details and temporal continuity from low-resolution video. Thanks to advancements in deep learning, deep video super-resolution has achieved promising results. However, frame artifacts and shakiness remain key challenges that need to be addressed for further improvements in video super-resolution. In particular, it is still worthwhile to research how to properly extract the spatiotemporal features from successive video frames and efficiently fuse them in order to achieve video super-resolution with rich intra-frame details and consistent inter-frame variations. In light of this, we propose to improve bi-directional recurrent network for video super-resolution with deformable motion alignment structure in this work to precisely estimate the offsets and provide sharper image frames. Experiments demonstrate that the proposed approach outperforms current state-of-the-art methods by 0.51%. The project code is available at https://github.com/hengliusky/IBRN_VSR.