Approximating Region Boundaries Based on Qualitative and Quantitative Information

Zhiguo Long, Michael Sioutis, Qingqian Li, Hua Meng, Heng-Chao Li

Abstract: Approximating regions is a topic that can have important applications in artificial intelligence whenever uncertain, incomplete, or inconsistent/contradictory spatial information is involved. This article devises a new method to generate region approximations based on rough qualitative direction and distance information. The main idea is first to give a suitability score to smaller regions, cells, that are obtained by partitioning the area of interest, and then to identify candidates to form an approximation by evaluating score contribution ratios under a certain threshold. This article designs a novel mechanism that compares the actual information of cells with the provided rough information in order to calculate suitability scores, and proposes to exploit a regressor model that can predict a threshold given certain suitability scores. Experimental results show that, given a good threshold, this new method can approximate target regions effectively, and that good thresholds can be reliably obtained through a trained regressor.