A line matching method based on local and global appearance

This paper proposes a line matching method based on both local neighborhood gradient and global structure information of lines. Firstly we generate an initial set of line segment correspondences using our improved Mean Standard Deviation Line Descriptor (MSLD). Then candidate matches those violate the global topological structure of lines are removed to eliminate wrong matches. Finally iterative topological filter is used to search for more matches, whilst global angle constrains are implemented to get rid of wrong matches and make the algorithm more efficient. Experiments show that the proposed method is highly robust under the conditions of heavy illumination change, rotation, image blur, viewpoint change, scale change, etc. Comparisons on image database demonstrate that our method greatly outperforms the state-of-art methods in matching accuracy and efficiency.