Matching of Polygon Objects by Optimizing Geometric Criteria

Despite the semantic criteria, geometric criteria have different performances on polygon feature matching in different vector datasets. By using these criteria for measuring the similarity of two polygons in all matchings, the same results would not have been obtained. To achieve the best matching results, the determination of optimal geometric criteria for each dataset is considered necessary. In previous research, the most used geometric criteria are the overlap area between two features, the Euclidian distance between two features, the orientation difference of two features, and the shape similarity of two features. In addition to determining the impact factor of each criterion in the best result, the best geometric criteria combination should be specified. In this study, unlike previous studies which have considered object matching as a unique issue in all datasets, objects matching is considered as a separate issue in each dataset and by converting the problem as an optimization problem, an approach is proposed to define optimal weights of criteria for different datasets using a genetic algorithm. In each dataset, corresponding best weights have distinguished that lead to the best matching result. To evaluate the proposed approach, a variety of spatial datasets of residential buildings have been used including a part of Bandar Abbas city in 1:25000, 1:50000, and 1:100000 scales; a part of district 6 of Tehran city in 1:25000 and 1:50000 scales; and a part of Rasht city in 1:25000, 1:50000, and 1:100000 scales. The results showed that the proposed approach has done a good performance in both polygon feature matching and identifying six corresponding relationship classes in all study areas. Moreover, matching results have been improved by an average of 28.61% compared to the case where all criteria are considered with equal weights and an average of 9.13% compared to the case that criteria are assessed according to expert opinions.