Scholarlink

<p>Geometric model fitting is a fundamental task in computer vision, often challenged by data corrupted with numerous outliers. To address this, we propose LGSPS (Local-to-Global Spatial Proximity Sampling), a robust sampling algorithm that leverages a local-to-global proximity measure to guide minimal subset selection. By constructing neighborhood sets based on true inliers and exploiting their local-to-global spatial relationships, LGSPS effectively suppresses structurally inconsistent outliers and generates minimal subsets that lead to reliable model hypotheses. Experimental results on single- and multi-structure real images demonstrate that LGSPS consistently outperforms state-of-the-art fitting methods, confirming its robustness and practical effectiveness. The source code is available at:<a href="https://github.com/lanzizi166/Robust-Local-to-Global-Spatial-Proximity-Sampling-for-Geometric-Model-Fitting">https://github.com/lanzizi166/Robust-Local-to-Global-Spatial-Proximity-Sampling-for-Geometric-Model-Fitting</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

Robust local-to-global spatial proximity sampling for geometric model fitting

  • 期刊论文

  • 发表日期：

    2026年7月1日

查看全文
  • Shaobo Kang,
  • Mingzhi Yang,
  • Rongwei Xu,
  • Juncheng Wang

摘要

Geometric model fitting is a fundamental task in computer vision, often challenged by data corrupted with numerous outliers. To address this, we propose LGSPS (Local-to-Global Spatial Proximity Sampling), a robust sampling algorithm that leverages a local-to-global proximity measure to guide minimal subset selection. By constructing neighborhood sets based on true inliers and exploiting their local-to-global spatial relationships, LGSPS effectively suppresses structurally inconsistent outliers and generates minimal subsets that lead to reliable model hypotheses. Experimental results on single- and multi-structure real images demonstrate that LGSPS consistently outperforms state-of-the-art fitting methods, confirming its robustness and practical effectiveness. The source code is available at:https://github.com/lanzizi166/Robust-Local-to-Global-Spatial-Proximity-Sampling-for-Geometric-Model-Fitting.