î SUltof the fact 
in this 
? 82)firs > propose 
me N. Let the 
v Motion 2 
nes s” in the space 
n Ae context of a 
“PU 1 *]. ^ 
i® 1 *. averaged 
se ' of all training 
^ire segmentation 
O' and parameter 
iew ), we obtained 
fware packages: 
and ASTRO 
and export the 
at. In total, 150 
, smoothness and 
1,0.3,0.5,0.7, 
ing the resultant 
-e computed the 
using JTS 
and GeoTools 
s and trucks) as 
ig rooftops for a 
ining object sets 
we computed 
ach combination 
I examined the 
objects inland 
lentation results 
alts are reported 
training objfl* 
ri-A tahflVlOf 01 
ter variation 
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008 
a. Vehicles, average of all segments in Y* 
c. Buildings, average of all segments in Y* 
e. Trees, average of all segments in Y* 
overSegmentation 
b. Vehicles, average of training objects 
0.4 0.6 0.8 
overSegmentation 
d. Buildings, average of training objects 
0.2 0.4 0.6 0.8 
overSegmentation 
f. Trees, average of training objects 
g. Average of all segments in Y* for all 
training objects 
Figure 1. The segmentation results when averaged overly E Yf, 
V if. ASTRO = □, eCognition = ♦. 
I "nringwlÉil