Quality assessment of building extraction from remote sensing imagery

An automatic quality assessment of extracted buildings from remote sensing imagery is needed to evaluate extraction algorithms, or to support change detection. In this paper, four commonly used measures are compared to the newly proposed metric for comparison of polygons and line segments (PoLiS). The extracted polygons are compared to the reference polygons and the quality measures are computed for each pair. The symmetric measures, i.e. quality rate and PoLiS, estimate overall dissimilarity between polygons, whereas i.e. the root mean square error (RMSE) of the distances between the polygon vertices, completeness, and correctness, are not symmetric and should be therefore used for applications like change detection. The variability of the measures is assessed according to the area of the reference buildings. The variability is higher for the category of larger buildings, where the building polygon complexity is larger.