联合共线约束与匹配冗余的组直线匹配结果检核算法

针对“一对多”“多对一”“多对多”直线匹配结果难以检核的瓶颈问题,提出一种共线约束与匹配冗余的组直线匹配结果检核算法。该算法在已有组直线匹配结果基础上,首先将结果中的每对同名直线组分裂为两对同名单直线,分别建立直线组、单直线同名对应关系矩阵,矩阵的行、列号分别对应参考影像、搜索影像上直线组索引或单直线索引,矩阵元素值可用于记录其行、列号所表示对应关系出现的数目、特征相似性系数等多源信息;在此基础上,以同名直线关系矩阵为基础,提取局部关系矩阵,建立“一对多”“多对一”“多对多”匹配结果间的关联;然后基于局部关系矩阵,结合共线约束、匹配冗余、特征相似性系数对结果进行检核,剔除错误匹配,保留正确匹配;最后对结果中的“多”直线进行拟合得到“一对一”的同名直线。选取具有典型纹理特征的航空影像和近景影像进行组直线匹配结果检核试验,获得了可靠的同名直线,验证了本文算法的有效性。

A checking algorithm for pair-wise line matching based on collinearity constraint and matching redundancy

A checking algorithm for pair-wise line matching is proposed to solve the problems of one-to-many, many-to-one and many-to-many in line segment matching by combining collinearity constraint and matching redundancy. The proposed method is performed on pair-wise line matches generated from existing method. Firstly, individual line segment matches are obtained based on pair-wise line matches. A relation matrix is constructed for pair-wise line matches and individual line segment matches, respectively. In each relation matrix, row and column numbers correspond to the indexes of matching primitives on the reference and searching images respectively. Each matrix element can be used to record the number of matches, feature similarity or some other multi-source information corresponding to the current row and column. Secondly, the corresponding relation of one-to-many, many-to-one and many-to-many matches is established on the basis of local relation matrix which is extracted from the aforementioned relation matrix. Then, the local relation matrix is combined with collinearity constraint, matching redundancy and feature similarity to identify and eliminate outliers in the matching results. Finally, the breaking lines in the results are fitted to produce one-to-one line segment correspondences. The proposed algorithm is evaluated on digital aerial images and close-images with typical texture features. The experimental results demonstrate the effectiveness of the proposed algorithm in pair-wise line matching result checking.