Scholarlink

<p>Local feature extraction and matching has lately attracted increasing attention due to its widely application, especially in real-time automated systems. However, an overall survey that discusses local feature-based image matching methods in detail is insufficient, and the related datasets suffer from a deficiency, especially for abnormal weather conditions. To address aforementioned problems, we first carry out a comprehensive review on existing local feature extraction and matching methods including handcrafted and learning-based methods, as well as related datasets and benchmarks. Second, a novel dataset with scenes of abnormal weather, named synthetic rain and fog for image matching (SRFIM), is proposed to cover the various scenes for local feature research. Finally, the robustness of 15 state-of-the-art approaches in different scenarios, encompassing changes in illumination, viewpoint, rain, and fog, is evaluated and analyzed. The experimental results demonstrate that learning-based methods are competent to deal with scenes of illumination and fog. However, in rainy scenes, handcrafted methods with specific designs outperform some learning-based methods at low pixel error thresholds. This finding highlights how the lack of abnormal weather datasets may constrain the generalization capability of deep learning methods. And for changes in viewpoint, two types of methods are both little competitive. Through applying outlier rejection techniques, the performance gap between handcrafted and learning-based methods in robustness evaluation become insignificant. The dataset and synthesis algorithms are publicly available at <a href="https://github.com/TakeoffC/SRFIM">https://github.com/TakeoffC/SRFIM</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

Local feature-based image matching: a comprehensive review and robustness evaluation

  • 期刊论文

  • 发表日期：

    2025年12月12日

查看全文
  • Houqin Bian,
  • Qifei Chen,
  • Haolin Zhang,
  • Lunming Qin,
  • Haoyang Cui,
  • Xi Wang

摘要

Local feature extraction and matching has lately attracted increasing attention due to its widely application, especially in real-time automated systems. However, an overall survey that discusses local feature-based image matching methods in detail is insufficient, and the related datasets suffer from a deficiency, especially for abnormal weather conditions. To address aforementioned problems, we first carry out a comprehensive review on existing local feature extraction and matching methods including handcrafted and learning-based methods, as well as related datasets and benchmarks. Second, a novel dataset with scenes of abnormal weather, named synthetic rain and fog for image matching (SRFIM), is proposed to cover the various scenes for local feature research. Finally, the robustness of 15 state-of-the-art approaches in different scenarios, encompassing changes in illumination, viewpoint, rain, and fog, is evaluated and analyzed. The experimental results demonstrate that learning-based methods are competent to deal with scenes of illumination and fog. However, in rainy scenes, handcrafted methods with specific designs outperform some learning-based methods at low pixel error thresholds. This finding highlights how the lack of abnormal weather datasets may constrain the generalization capability of deep learning methods. And for changes in viewpoint, two types of methods are both little competitive. Through applying outlier rejection techniques, the performance gap between handcrafted and learning-based methods in robustness evaluation become insignificant. The dataset and synthesis algorithms are publicly available at https://github.com/TakeoffC/SRFIM.