in, 
itial 
as: 
nts, 
les. 
een 
wal 
the 
lose 
and 
ZZy 
sis, 
ical 
ises 
iper 
d is 
ines 
e Is 
york 
city 
| by 
note 
10n- 
d in 
ship 
is in 
g an 
1g a 
[AX 
1ZZY 
| be 
the 
on 
ined 
=25 
nds. 
IZZy 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXX V, Part B3. Istanbul 2004 
Membership Functions Plot 
  
      
08 
06. 
04 
02 
0 
NE 
E S 
2n 2, 50 NN 
7 2 S Ww "3 aM a 
IC weal ^, 
Figure 1. Defined classes using MFs of bands 1 and 2 
  
| Input Multispectral Image 
  
  
Sampling from road surface 
to obtain its u and © 
  
  
Membership function definition 
(Gaussian Fuzzification) 
  
  
[Formation of 125 fuzzy classes | 
v 
| MIN and MAX operations ) 
Ÿ 
| Defuzzification ] 
  
  
  
Figure 2. Implementation flowchart in fuzzy step 
A fuzzy class c in band b is defined by f, (x,) where x, is 
the grey level of the pixel in band b. 4, , is the mean value of 
class c in band b. o, is the standard deviation of class c in 
b.c 
band bh. The pixel vector X in the B-dimensional space is 
(Melghani et al., 2000): 
X={x x, Kx, K, x] 
Cu = Hoch Y. [36 
ie es E (1) 
246 
be 
f, ,) - exp(- 
Vx, e[0255]. V f£.(x,)=] 
For the hard classification, first a MIN operation is 
applied on each column of the matrix / Then a MAX operation 
will be performed on these elements to obtain the element with 
the highest value (fuzzy output) and the corresponding class of 
that element will be considered as associated fuzzy class to that 
pixel. As mentioned before, there are 125 hypothetic classes. 
Among these classes, only 64" class is road (this depends on 
the order of the classes in the matrix f). Other classes will be 
765 
considered as non-road. In this manner, the image is segmented 
into road and non-road segment. 
2.2 Stage2: Mathematical Morphology 
Line based methods for automatic road network extraction from 
high resolution images involves edge-line detection, threshold 
selection, grouping and road linking. The difficulties arise when 
threshold selection and linking based on conditions such as 
proximity, orientation and geometrical constraints. With the 
complexity in the image due to the occlusions and difference of 
materials on both sides of road, these conditions normally are 
not satisfied specially for Iranian roads. This makes line based 
methods less effective for high resolution images. The approach 
used in the second stage is shown in figure 3 which is nearly 
close to the approach used by Zhang et al., (1999). 
        
   
  
  
Remove noises 
Remove small paths 
Jecond Trivial opening 
Fuzzy classification 
Segmentation 
Granulometry 
Trivial opening 
[Output center line of the road network | 
     
  
      
    
  
     
.Josing sma 
= 
  
   
gaps 
  
  
Figure 3. Block diagram used in morphology approach 
Trivial opening is defined by Serra and Vincent (1993). It 
provides a practical mean of object detection and identification. 
It does not affect the shape and size of the objects of interest. 
Let X be a collection of connected pixels of objects where Xi is 
an object in X. Then, with a criterion 7: 
X (i), if X(i) satisfies criterian T 
Q, Otherwise 
V -Qemi x (2) 
recon, (N= YEE § 4 
n times 
Trivial Opening = 
recon, (Y) is the geodesic dilation of order n and J,, is the 
elementary geodesic dilation. Assuming a pixel Y is in X7 then 
Xi is reconstructed from VY by iterating the elementary geodesic 
dilation until the whole object is covered. Figure 4 
demonstrates the reconstruction of an object by morphological 
reconstruction. 
The above mentioned trivial opening can be used to measure 
the size and shape of objects in an image. The opened images 
are compared with the original image to generate measures with 
respect to different size of structure element but with same 
shape. These measures can be used as shape and size signature 
of the original image (Granulometry) and can be plotted as a 
pattern spectrum.