International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B-YF. Istanbul 2004 
  
Sin 
y - = wy ? 
* . * 
» ^ e mE e y A 
^A KR » - wir 
= , 3 " “in 
> LS n. f» s 
- + 4 ® mm - " - n » 
T. B. pow s 
i 4^ >» - ye i 
1.r i 8 ^ - 
bt ? 2 S 2 . 
^ $ ep MS ew ew 7 IN 
» 1 si 
/ à £t se A - erg : 
zen * * 9 vat Le & 
: 3 ‘ 
2 4 “À t, x 
e = NES ¥ 
\ E mate ae 7 
\ $ [2] 
\ » 3 
- nose wol * > we ute! 
yy 3 en - 2 7 
SS 
> B b " - 
‚ mn a F » 4 
z ". 
es T v : 
- . x 
' ~ a vn. s — E 
* c» 55 - " 
18 on s » 
9 > + 2e 4o y7X ES 
MR. - 
: » 8. s zv . ™ 
EP > 274, NS 
2 gi 00e hos > 
* NT y dee . » 44 
A ^ æ T + +s T 
i : 7 e 
« cris - ma br +» a. se 
Fig.3 Light objects sensible to structure element N, 
T 
X 
E 
: & 
  
  
Fig.4 Obj 
ects insensible to the structure element sequence 
   
a t, cod! 
a emi. v Xi 
    
rm 3 
» dhe * 
  
Fig.7 Result obtained by the proposed method in this paper 
The results of multi-scale segmentation to remotely sensed 
image are shown in Fig. 7. In this resultant image, the buildings, 
water, grassland and roads can be basically distinguished. Fig. 8 
is the result of watershed segmentation, which is calculated by 
Matlab. It can be seen that over segmentation and boundary 
pixels spoiled the results. Compared with Fig. 8, 
the 
segmentation approach proposed in this paper is better than 
50 
Fig.8 Result obtained by the watershed segmentation 
classical watershed segmentation, because the algorithm here 
avoids the problems of over segmentation and boundary pixels 
by obtaining morphology features of objects. Different types of 
objects can fall into different resultant images. In this way, the 
original image can be effectively segmented. 
at