Relaxation Labelling Matching for Multi-scale Residential Datasets Based on Neighboring Patterns

This paper proposes a relaxation labelling matching approach for multi-scale residential datasets based on neighboring patterns. Firstly, we detect the candidate matching objects and neighboring patterns by buffering analysis and spatial neighboring relations. Secondly, the geometric similarities of candidate matching objects or neighboring patterns are calculated to initialize the matching matrix that contains 1:1, 1:M and M:N relations. After that, the contextual information of neighborhood objects or patterns are explored to heuristically update the matching matrix to achieve a global consistency. The matching pairs with maximum probabilities are finally selected after context consistency detection. The experimental results and contrast analysis show that our method obtains high correct matching rates, efficiently overcomes the problems of shape homogeneity and uneven deviation, and can correctly identify complex 1:M and M:N matching objects in multi-scale residential datasets.