基于描述子与几何约束的直线段匹配  

Line segment matching based on descriptor and geometric constraints

作　　者：武盟盟 焦伟利[1] 龙腾飞[1] 

机构地区：[1]中国科学院遥感与数字地球研究所,北京100094 [2]中国科学院大学,北京100049

出　　处：《计算机工程与设计》2014年第6期1983-1987,1998,共6页Computer Engineering and Design

基　　金：国家自然科学基金项目(61271013);国家科技支撑计划基金项目(2012BAH27B05)

摘　　要：提出了基于描述子与几何约束的直线段匹配方法。采用LSD(line segment detection)算法提取直线特征并用LBD(line band descriptor)描述子表示直线,通过计算2幅影像中直线描述子之间的距离得到N(N>3)对描述子距离最小的匹配直线对;估计影像间的变换模型参数,以变换模型约束下直线段中点间的距离和直线段间的距离作为相似性测度实现直线段匹配。与现有的直线段匹配算法进行对比,比对结果表明,对于真实的遥感影像及仿射变换、旋转、噪声、光照变化等仿真遥感影像,该方法均得到了更多的匹配直线段对,且耗时更少、匹配准确率高。A line segment matching method based on descriptor and geometric constraints was proposed.Firstly,the algorithm LSD （line segment detection） was used to extract line features and then LBD （line band descriptor） was used to represent every line feature.Secondly,the distances between the line descriptors of the two images were caloalated to obtain the N minimal distances,namely N pairs of matched line segments where N was bigger than three,and then the transform model parameters are estimated by the N pairs of matched line segments.Lastly,the distance between midpoints of two line segments and the distance between two line segments were used as constraints to implement line segment matching.Compared with existing algorithm,the experimental results show that the method proposed could obtain more matches and had high matching accuracy with less time in true remote sensing images and simulation remote sensing images,including affine transform,rotation,noise and illumination.

关 键 词：直线特征 描述子 几何约束 直线段匹配 遥感影像 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]