摘要
针对弱纹理目标匹配问题,提出了一种基于直线局部邻域梯度信息和全局结构信息的直线匹配算法:对均值标准差直线描述符进行改进用于初始匹配;利用直线间的全局拓扑结构滤除误匹配;利用迭代拓扑滤波寻找更多的匹配,同时引入全局角度约束提高算法效率并进一步滤除错误匹配。实验表明,在光照变化、图像旋转、图像模糊、尺度变换、视点变化等条件下,该方法都具有很强的鲁棒性,并在匹配效率和准确度上优于现有的两种比较流行的方法。
In view of the matching problem of low texture objects,a line matching algorithm based on local neighboring gradient information and global structural information was proposed. The mean standard deviation line descriptor was redesigned to get the initial matching; the global topological structure between lines was used to get rid of wrong matches; more matches were achieved by utilizing the iterative topological filter.Moreover,global angle constrains were implemented to make the algorithm more efficient and to further remove wrong matches. Experiments show that the proposed algorithm is highly robust under various image changes including heavy illumination change,image rotation,image blur,scale change as well as viewpoint change and is much better than the other two popular methods in terms of matching accuracy and efficiency.
出处
《国防科技大学学报》
EI
CAS
CSCD
北大核心
2014年第6期25-30,共6页
Journal of National University of Defense Technology
基金
国家重点基础研究发展计划(973计划)资助项目(2013CB733100)
关键词
直线检测
直线描述
直线匹配
拓扑约束
line detection
line descriptor
line matching
topological constraint
