AN EDGE DETECTOR BASED ON WIDE-NARROW MORPHOLOGICAL OPERATIONS 
OF SATELLITE REMOTE SENSING IMAGES 
Makoto KAWAMURA, Toyohashi University of Technology, JAPAN 
Yuji TSUJIKO, Fukui National College of Technology, JAPAN 
Sanath JAYAMANNA, Toyohashi University of Technology, JAPAN 
Commission Il!, Working Group 3 
KEY WORDS: Remote Sensing, Classification, Extraction, Algorithm, Edge, Landsat 
ABSTRACT 
In a classification of satellite remote sensing data, spectral distribution in a feature space is usually used. 
However, MIXELs at class boundaries cause miss-classification results. To solve this problem many 
researchers have carried out increasing the feature space dimensions by giving additional information. In 
this study, we describes the method creating additional information mentioned above. In particular, a 
new Wide-Narrow Morphological Edge Detection (WNED) algorithm to make edge information is 
introduced. WNED differs from previous morphological edge detectors in that it manipulates two 
conventional minimum-based operations in the target domain at the same time. As the results of a case 
study for Landsat TM data it is found that WNED algorithm is effective to extract edge information clearly 
and it can control the detection of the spurious edge information. 
1. INTRODUCTION 
Remote sensing technology has contributed in 
assisting to make accurate maps timely, widely 
and economically. A large number of studies have 
introduced the effectiveness of satellite remote 
sensing data and ifs applicabilities for the 
monitoring. In the land cover classification, 
however, mixed pixels (mixel) which are laid 
among some categories produce the 
miss-classification outpts. To extract or to avoid 
these pixels, many researchers have tried to detect 
edge pixels using conventional segmentation 
methods such as convolutional filterings. Recently, 
there are some cases using mathematical 
morphology to execute the segmentation 
(Kawamura, 1994 and 1995)). This study also 
describes the segmentation method in terms of 
edge detection to improve the classification 
accuracy using mathematical morphology. In 
particular, a new algorithm to make accurate and 
clear edge information is introduced. 
Theoretical background of mathematical 
morphology was established at the Centre de 
Morphologie Mathematique in France, in the mid 
1970's (Matheron, 1975) and extended to 
application for image processing (Haralick, 1987). 
It is also well-known that image shape features 
such as edges, fillets, holes and skeletons can be 
obtained by combining morphological 
fundamental operations and structuring elements. 
In this study, the combination of morphological 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996 
fundamental operations and structuring elements 
is focused. 
2. MORPHOLOGICAL OPERATION 
Morphological operations are classified in a binary 
morphology and a gray-scale morphology. Usually 
satellite remote sensing data are given as 
gray-scale image. Therefore this analysis describes 
the gray-scale morphological operations. The 
gray-scale morphological operations can be 
defined as follows. Let f(x) and k(x) be 1-dimesional 
gray-tone functions of coordinate x, where f(x) is 
the original remote sensing image, and k(x) is the 
operator (filter) called structuring element. Let E 
represent Euclidean space. Then f:F?E and k:K—E. 
The dilation and the erosion operations are defined 
as follows: 
dilation : (f®k) (x)=max{f(x-z)+k(z)}=d(x) 
VzC€K, xzC€F (1) 
erosion : (fO k) (x) »min(f(x*z)-k(z))&a e(x). 
VzCK, xvz€F (2) 
The opening and closing operations can be 
defined in terms of dilation and erosion operations 
as follows: 
886 
  
    
    
   
  
  
  
  
  
  
  
  
  
  
  
  
   
    
   
  
     
   
     
    
    
    
   
    
    
     
   
    
     
     
     
   
   
   
  
  
   
WC) CE TOO DOT 
— — 0) 75