methods, considering whether the roofs were partially or fully
extracted or not. “Over-segmentation” refers to the situation
where a house is broken into several smaller segments. “Un-
der-segmentation” refers to the situation where two or more
houses are combined into one. There are totally 108 houses
in Dataset 1. Our method extracted all of them. There was one
over-segmentation and one under-segmentation. The general
marker-watershed method extracted 119 houses, whereas
eight houses were over segmented, and two houses were un-
der segmented. Pesaresi’s method required further processing
to extract the final house masks, so only 16 under-segmented
houses counted. It shows that Pesaresi’s method often results
(a)
(b)
Figure 6. Markers and models for the high-density residential area of Dataset 2: (a) DSM, and (b) models of buildings.
(a)
(b)
Figure 7. Markers and models for the high-density residential area of Dataset 3: (a) DSM, and (b) models of buildings.
26
January 2016
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
