€ 
Figure 3 Original image (f(x)) 
    
  
    
     
   
  
  
  
  
  
  
  
    
  
    
   
  
   
     
   
    
    
   
   
     
    
    
   
  
WNED(x)=fo-max{DE, EE} (9) 
where fo is output of wide range detection. In this 
study, we use: 
fo=max{d(x)-o(x), c(x)-e(x)}. (10) 
4. CASE STUDY 
4.1 Data 
NVI image given by the near infrared band (band 
4) and the visible image band (band 3) of Landsat 
TM purchased from RESTEC, JAPAN is used as the 
original image. It is for path=109 and row=36 which 
covers the center of Nagoya, Japan. Particularly 
100X 100 pixels surroundings of Kiso river are 
selected. Figure 3 shows the original image. The 
target edge pixels are boundaries of land and river. 
4.2 Selection of structuring element 
There are many types of structuring elements in the 
previous works (Maragos, 1987). This analysis 
provides four types of structuring elements as 
shown in Figure 4. In this paper, the cases applying 
3X3 (Figure 4(a))are introduced. 
4.3 Results 
Figure 5 all show the results by morphological edge 
detectors. (a) and (b) show the results of applying 
the DE and EE with structuring elements (k(x)) of 
size 3X 3 respectively. Many different edge 
intensities are found in these two images. (c) in the 
same figure show the minimum of DE and EE which 
is a similar mannar to BMM and ATM. In the figure, 
  
  
(b) line direction 
(d) slant direction 
(a) rectangular(basic) 
structuring element 
  
A718 
Ge 
structuring element 
(c) column direction 
structuring element 
        
structuring element 1. &zx 
(e) slant direction 
structuring element 2 
  
  
888 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996 
Figure 4 Structuring elements (k(x)) 
many undesiable pixels are extracted. In the case 
of increasing the structuring element sizes, the 
extracted pixels of (c) were expanded according 
to the expansion of structuring elements. WNED 
outputs are shown in Figure 5(d). In comparison 
with other outputs, it is found that WNED can 
extract edge clearly and control the spurious edge 
detection. For increasing structuring element sizes, 
WNED did not expand the edge pixels comparing 
with the minimum based edge detection. 
5. CONCLUSION 
It is found that WNED algorithm is effective in case 
of detecting rapid change of gray-tone functions 
such as boundaries of the river and it can control 
the spurious edge detection. Proposed algorithm is 
also useful for the unknown target, because it 
doesn't expand the edge pixels regardless of 
increasing structuring element sizes. 
By using this method it is considered that it is able 
to use monitorings the disaster such as a flood. 
REFERENCES 
Feehs R. and Ace G., 1987. Multidimensional edge 
detection. Proc. SPIE, Visual Commun. Image 
Processing Il, Vol.845, pp. 285-292. 
Haralick R., Sternberg .S. and Zhuang .X., 1987. 
Image analysis using mathematical morphology. 
IEEE Trans. . Pattern Anal. | Machine Intell, 
Vol.PATM-9. 
Kawamura M. and TSUJIKO Y., 1994. Multispectral 
classification of Landsat IM data. using. .a