d by Zhang and 
binary image. In 
The method con- 
> steps applied to 
where a contour 
d having one 8- 
nce to the 8- 
gure 2), the first 
1 if the following 
©) 
ero neighbors of 
(10) 
tions in the order 
le, N(pl)=4 and 
yindow 
| b in equation 
| (b) remain the 
iged to, 
(11) 
el in the binary 
ough (d) are vio- 
t changed. If all 
for deletion. 
ng edge points is 
to analyze the characteristics of pixels in a small neigh- 
borhood (e.g.3x3 or 5x5) about every point (x,y) in an 
image that has undergone an edge-detection process. All 
points that are similar are linked, thus forming the 
boundary of pixels that share some common properties. 
The two principle properties used for establishing simi- 
larity of edge pixel in such analysis are : 
1- The strength of the response of the gradient operator 
used to produce the edge pixel, and 
2- The direction of the gradient. 
The first property is given by the value of 
Gmag[f(x,y)] . Thus it is said that an edge pixel with co- 
ordinates (x,y) and in the predefined neighborhood of 
(x,y), is similar in magnitude to the pixel at (x,y) if 
| Gmag [fGx.y) ] - Gmag [xy IST (12) 
Also it is said that an edge pixel at (x,y) in the prede- 
fined neighborhood of (x,y) has an angle similar to the 
pixel at (x,y) if 
oxy) -axy) |S A (13) 
where A is an angle threshold. The value of Gmag(.) 
and a (.) are calculated using the following equations . 
Gmag[fx,y)] = [Gx? + Gy2112 (14) 
o. tan 1 (Gy/ Gx) (15) 
Based on the concepts given above, a point in the prede- 
fined neighborhood of (x,y) is linked to the pixel at (x.y) 
if both the magnitude and direction criteria are satisfied. 
This process is repeated for every location in the image. 
3. Developed method 
A developed method for extracting straight line seg- 
ments is presented here. This method starts first with 
edge detection process. Figure 3 shows the results of 
different edge detectors with automatic thresholding 
technique. The Sobel operator is used since it provides 
clear edges and is not very much affected by pulsive 
noise. 
Figure 5 shows edge maps using the thresholding tech- 
nique mentioned above. 
The automatic thresholding technique is done by divid- 
ing the gradient image g(x,y) into subimages with spe- 
cific size, say (zxz). For each sub-image the average in- 
tensity (av) is calculated and each pixel within this 
subimage is compared with this average. An edge ele- 
ment is said to be present if the value of the pixel inten- 
sity is greater than the calculated average multiplied by 
a tuning factor(B). The above process can be implement- 
ed as follows: 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996 
  
   
  
   
   
   
  
   
   
  
  
  
   
   
   
  
   
     
   
    
   
   
    
  
   
   
   
    
   
   
   
    
     
   
   
   
  
  
   
     
   
     
   
      
Z-1 2-1 
av(x,y) -— > g (x+i ‚y+j) (16) 
i=0 j=0 
1 if g(xH,y+]) > B.av (xy) 
e («+i,y+]) = (17) 
0 otherwise 
where x,y = 0,2,2z .....(N-z), 
N is the image dimensions , and 
B is the tuning factor. 
The value of B dictates the resultant edges. In order to 
obtain a large number of edges the value of b is set less 
than (1) and if only the strong edges are required the 
value of (B) is set greater than (1). Normally the value of 
B is almost always equal to (1). 
The resultant edge image is coded to reduce the storage 
space required and speed up the processing. The edge 
element number corresponds to each direction. The val- 
ue of this integer depends on the direction of the edge 
and the number of quantization levels. 
The coded image is obtained according to the following 
criteria :- 
if (Omas I) 
if (Gmag > T) and (0 £ 0 mod 180 < 45) 
if (Gmag > T) and (45 £ & mod 180 < 90) 
if (Gmag > T) and (90 < @ mod 180 < 135) 
if (Gmag > T) and (135 < & mod 180 < 180) 
e (x,y)= 
SV Di = © 
(18) 
From equation (18) it is clear that (0) represents the 
background and the rest of the integers represents the 
edge elements. 
In order to obtain an edge image containing this edges 
and only the dominant edges. One can merge the two 
edge images properties by multiplying them with each 
other (( considering that (0) represents background and 
(1) represent edge element )) as shown in Figure (6) . 
Sometimes this process is used instead if using the thin- 
ning process. 
While the use of thinning process is more effective way 
to obtain thin edges. 
The thinning algorithm that have been discussed in sec- 
tion(2.2.2) are applied on the edges obtained form the 
edge detection and the results are shown in Figure 7. 
The technique developed by (Nevatia and Babu, 1980) 
is used here. This method was chosen because it pro- 
vides a thin and connected edges . A few minor modifi- 
31