A new descriptor for improving geometric-based matching of linear objects on multi-scale datasets

Geometric-based matching identifies corresponding objects in multi-representation datasets using geometric and topological properties. The criteria used in the matching process encounter ambiguities in some cases, which cause to reducing accuracy of matching in identifying corresponding paired objects. This paper aims to present a descriptor, named as Rotary summation based on orientation and distance, which can be used as a criterion in linear object matching to improve the matching results along with other geometric and topological criteria. The proposed descriptor, which is based on distance and orientation, identifies undetectable corresponding objects through other criteria by taking account the spatial relations between the objects and extracted landmarks in datasets of different scales and sources. The efficiency of the proposed descriptor in improvement of the accuracy of matching is evaluated by six datasets with different scales from two different areas. The results show improvements in matching by considering the proposed descriptor such that the F score value increases on average by 5.48% in the studying datasets.