tellites, 
| areas. 
ject in 
sh both 
e study 
inalysis 
.ONOS 
ti-scale 
lysis to 
f the river 
ion result 
fication 
100d 
image 
ry 
' similar 
similar 
  
bamboo 
  
rom 
ry 
tify 
aly 
be used 
num 
n 
one 
y be 
ication 
at 
Tom 
nagery 
od 
| area 18 
E 
1ade 
  
  
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004 
Ground resolution of IKONOS data is smaller than the average 
patch sizes of the vegetation classes of this study, and pixel 
values alone may not provide enough information for 
distinguishing image objects corresponding to the instances of 
vegetation classes, although pixel-based classifiers such as 
maximum-likelihood classifiers provide useful vegetation 
information such as vegetation density and distribution in 
individual patches of vegetation classes. On the other hand, 
object-based classifiers can utilize contextual information as 
well as spectral information for image classification, and are 
suitable for delineating image objects corresponding to the 
vegetation patches on IKONOS imagery. They group 
contiguous pixels with similar pixel values into image objects 
and label them according to contextual information such as 
spatial relationships among image objects as well as spectral 
and textural properties. 
3. STUDY AREA AND DATA 
In this study, a trial of the hybrid vegetation mapping 
from actual IKONOS data was conducted. The lower part 
of the Nivodo river watershed area located in the south- 
western part of Japan was chosen for the study area. The 
land use of the study area mainly consists of natural 
forests and agricultural fields (rice fields, vegetable fields, 
  
  
  
green houses, etc.). 
The remote sensing data used in this study is the IKONOS pan- 
sharpen CIR imagery data produced from the panchromatic and 
multi-spectral IKONOS data taken on November 21, 2001 for 
the study area. 
4. IMAGE SEGMENTATION 
Object-based classification starts with the image segmentation 
process, which delineate image objects with relatively similar 
properties according to segmentation criteria. In this study, the 
segmentation algorithm developed by Baatz and Schäpe (2000), 
implemented in the e-cognition* * software, was used to 
conduct image segmentation of the IKONOS data with multiple 
scale parameters to see the correspondence between image 
objects and land cover instances including vegetation patches. 
The calculation parameters of the multi-scale segmentation used 
in the segmentation were "colour" (weight 0.8), "shape" 
(weight 0.2, further, the “shape” parameter consists of 
“smoothness” (weight 0.9) and “compactness” (weight 0.1)). 
Figure 1 shows some examples of the segmentation results at 
different scale parameters. These are at some intermediate steps 
of a region growing process from individual pixels. Figure 1 (b) 
  
Figure 1. Segmentation results with different scale parameters 
(a) IKONOS (* 3SI) image, (b) Scale parameter 20, (c) Scale parameter 80, (d) scale parameter 300 
287