Scholarlink

<p>Image registration holds a pivotal role in computer vision, yet large-deformation compensation presents a significant challenge in this domain due to the difficulty of finding precise matches under substantial deformations. Taking inspiration from the collective nature of animal migration, we propose a novel perspective on image registration by conceptualizing large-deformation compensation as a form of collective motion. In this perspective, we divide image pixels into leaders and flocks. The collective motion is driven by leaders through three components: route decision based on collectiveness quantification, route execution with collectiveness maintenance, and cascaded migration with collectiveness inheritance. These components are integrated into a collective cascaded migration (CCM) framework, which facilitates effective compensation for large-deformations. Our extensive experiments on six challenging datasets, encompassing five 2D datasets and one 3D dataset for large-deformation compensation, conclusively demonstrate the effectiveness of CCM in accurately compensating for large-scale deformations. The related code and data are available at <a href="https://github.com/GouZi2019/CCM.">https://github.com/GouZi2019/CCM.</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

Collective Migration-Inspired Large-Deformation Compensation for Nonrigid Image Registration

  • 期刊论文

  • 发表日期：

    2025年9月20日

查看全文
  • Dingkun Liu,
  • Danni Ai,
  • Hong Song,
  • Jingfan Fan,
  • Tianyu Fu,
  • Deqiang Xiao,
  • Yuanyuan Wang,
  • Yongtian Wang,
  • Jian Yang

摘要

Image registration holds a pivotal role in computer vision, yet large-deformation compensation presents a significant challenge in this domain due to the difficulty of finding precise matches under substantial deformations. Taking inspiration from the collective nature of animal migration, we propose a novel perspective on image registration by conceptualizing large-deformation compensation as a form of collective motion. In this perspective, we divide image pixels into leaders and flocks. The collective motion is driven by leaders through three components: route decision based on collectiveness quantification, route execution with collectiveness maintenance, and cascaded migration with collectiveness inheritance. These components are integrated into a collective cascaded migration (CCM) framework, which facilitates effective compensation for large-deformations. Our extensive experiments on six challenging datasets, encompassing five 2D datasets and one 3D dataset for large-deformation compensation, conclusively demonstrate the effectiveness of CCM in accurately compensating for large-scale deformations. The related code and data are available at https://github.com/GouZi2019/CCM.