重访弧支撑线段:一种高效高质量的椭圆检测
摘要
多年来许多椭圆检测算法涌现并被广泛研究,而在真实图像中准确且高效地检测椭圆这一关键问题仍具挑战性。本文提出一种面向工业应用的、基于弧支撑线段的有价值椭圆检测器,其同时达到高检测精度与高效率。为简化图像中复杂曲线同时保留凸性与极性等一般属性,提取弧支撑线段,这为椭圆的成功检测奠定基础。通过将潜在属于同一椭圆的弧支撑线段迭代且鲁棒地连接形成弧支撑组。随后,采用两种互补方法,即局部选择显著性更高的弧支撑组与全局搜索所有有效配对组,以快速拟合初始椭圆。接着,通过对初始椭圆的 5 维参数空间进行层次聚类构造椭圆候选集。最后,选取显著椭圆候选并依严格有效的验证精炼为检测结果。在三个公开数据集上的大量实验表明,相较 SOTA 方法,我们的方法取得了最佳的 F-measure 分数。源代码见 https://github.com/AlanLuSun/High-quality-ellipse-detection。
引用
@article{arxiv.1810.03243,
title = {Arc-support Line Segments Revisited: An Efficient and High-quality Ellipse Detection},
author = {Changsheng Lu and Siyu Xia and Ming Shao and Yun Fu},
journal= {arXiv preprint arXiv:1810.03243},
year = {2019}
}
备注
IEEE Transactions on Image Processing; The paper has been revised to a more appropriate title "Arc-support Line Segments Revisited: An Efficient and High-quality Ellipse Detection"; In addition, please contacts us if you want a higher quality pdf due to the limited uploading size of files in arXiv where the resolution of figures might be not very high