Premium
Conflating linear features using turning function distance: A new orientation‐sensitive similarity measure
Author(s) -
Lei Ting L.,
Wang Rongrong
Publication year - 2021
Publication title -
transactions in gis
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.721
H-Index - 63
eISSN - 1467-9671
pISSN - 1361-1682
DOI - 10.1111/tgis.12726
Subject(s) - artificial intelligence , similarity (geometry) , conflation , hausdorff distance , pattern recognition (psychology) , offset (computer science) , feature (linguistics) , computer science , metric (unit) , similarity measure , mathematics , position (finance) , feature matching , triangle inequality , computer vision , feature extraction , image (mathematics) , engineering , combinatorics , philosophy , linguistics , operations management , epistemology , finance , economics , programming language
Abstract Measuring the similarity between counterpart geospatial features is crucial in the effective conflation of spatial datasets from difference sources. This article proposes a new similarity metric called the “map turning function distance” (MTFD) for matching linear features such as roads based on the well‐known turning function (TF) distance in computer vision. The MTFD overcomes the limitations of the traditional TF distance, such as the inability to handle partial matches and insensitivity to differences in scale and rotation. In particular, the MTFD allows one to: (a) partially match a linear feature to a portion of a larger feature from a certain position of match; and (b) consider both the shape and orientation differences of polylines based on comparing their turning angles. In finding the best match position, we prove that the optimal position can be found among a finite set of positions on the target feature. We then combine the MTFD with widely used point‐offset distances such as the Hausdorff distance to form a composite similarity metric. Our experiments with real road datasets demonstrate that the new metric has greater discriminative power than traditional point‐offset‐based similarity measures, and significantly improves the precision of two tested conflation models.