基于多种几何约束的图像误匹配剔除算法  

A mismatch removing method based on multiple geometric constraints

作　　者：刘李漫[1] 谭龙雨 刘佳 LIU Liman;TAN Longyu;LIU Jia(College of Biomedical Engineering,South⁃Central University for Nationalities,Wuhan,430074,China)

机构地区：[1]中南民族大学生物医学工程学院

出　　处：《中南民族大学学报（自然科学版）》2020年第1期91-97,共7页Journal of South-Central University for Nationalities：Natural Science Edition

基　　金：国家自然科学基金资助项目(61976227);湖北省自然科学基金资助项目(2019CFB622);国防某重点实验室基金(6142113180202);央高校基本科研业务费专项资金资助项目(CZY17011)

摘　　要：尺度不变特征变换SIFT由于其良好的性能在图像匹配中具有广泛应用,但传统的SIFT特征匹配结果中仍然会包含不少的错误匹配.提出了一种基于多种几何约束的误匹配剔除算法,该方法利用匹配点对之间多种几何约束的组合来识别并剔除错误的匹配,具体包括方向角算法、邻域信息算法、点线距离算法和局部相似三角形算法.实验结果证明:这一基于组合策略的方法能有效地剔除错误匹配,并且极大限度地保留正确匹配.SIFT feature has been extensively used to image matching due to the good performance. However, many false matches are still induced in the traditional SIFT feature matching. A mismatch removing method based on multiple geometric constraints is proposed. This method combines multiple geometric constraints between match pairs to identify and discard false matches. Four sub-algorithms including the direction angle algorithm, the neighboring information algorithm, the point-line algorithm and the local similar triangle algorithm are combined in the proposed method. Experiments show that the combinatorial strategy can effectively remove false matches while preserving the most correct matches.

关 键 词：图像匹配 误匹配剔除 几何约束 组合策略 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]