Miranda-2020

Conference proceedings article

Current research has shown significant interest in spatio-temporal data. The acquisition of spatio-temporal data usually begins with the segmentation of the objects of interest from raw data, which are then simplified and represented as polygons (contours). However, the simplification is usually performed individually, i.e., one polygon at a time, without considering additional information that can be inferred by looking at the correspondences between the polygons obtained from consecutive snapshots. This can reduce the quality of polygon matching, as the simplification algorithm may choose to remove vertices that would be relevant for the matching and maintain other less relevant ones. This causes undesired situations like unmatched vertices and multiple matched vertices. This paper presents a new methodology for polygon simplification that operates on pairs of shapes. The aim is to reduce the occurrence of unmatched and multiple matched vertices, while maintaining relevant vertices for image representation. We evaluated our method on synthetic and real world data and performed an extensive comparative study with two well-known simplification algorithms. The results show that our method outperforms current simplification algorithms, as it reduces the amount of unmatched vertexes and of vertexes with multiple matches.