【摘要】 快速影像匹配是进行影像时间序列分析与飞行器导航的重要方法。本文对待匹配影像进行高斯低通滤波预处理时,运用影像的尺度空间表达思想对不同分辨率的基准影像和实时影像选择了相应的σ值进行卷积滤波处理,使基准影像和实时影像具有相近的分辨率,从而提高两影像所提取角点的重复率,使得影像的正确匹配概率得到提高。然后用基于影像几何结构分析的改进的快速角点探测算法进行了影像的角点提取;最后用本文提出的改进的鲁棒Hausdorff距离进行了基准影像和实时影像的匹配。实验证明,本文方法对影像噪声和灰度变化不敏感,具有抗影像尺度变化的能力。在基准影像和实时影像灰度差变化较大的情况下,　　收稿日期:2004 07 16;修回日期:2004 11 15基金项目:航天预研基金资助项目(2003)作者简介:安　如(1963 ),女,江苏淮安人,副教授,博士生,目前主要研究方向为数字影像处理与匹配。依然能取得较高的正确匹配概率。由于采用基于影像信息量评价的搜索策略和快速角点提取算法,匹配速度也较快。更多还原

【Abstract】 Fast image matching is an important method for aircraft navigation and motion analysis. A fast image matching method is adopted based on scale-space representation of image and a new robust Hausdorff distance. Firstly, preprocess of the reference image and real-time image uses Gaussian Low-filter with different σ according to their space resolution. Therefore, a high repeatability of the corners is obtained and registration probability is improved. Then, corners are detected quickly based on image geometric structure analysis. Finally, a novel robust Hausdorff distance is used to match reference image and real-time image. The experiment result shows that the method the paper suggested is robust to noise, illumination and scale changes. It can be obtainable for high registration probability at the condition of existing bigger gray level difference between reference image and real-time image.更多还原