International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004 
  
area image and the total area image. And the deviation of the 
residential area pixels presents the grey distribution consistency. 
If the deviation is low, the grey distribution consistency is high 
for the threshold calculation. 
       
3a Original Image 
    
3c Gauss Blur Deviation Image 
3b Gauss Blur Image 
Fig 3 Comparison of Gauss Blur Image 
Table 2 Comparison of the statistic data in the original image and gauss blur images 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
Feature Value Max | Min Dev Aver | Entropy | Dif. Of Dev p of Dif. Of Entropy 
L6 Total Area 255 0 41.27 128.7 7.30 ; 
Original Image. metal Area | 255 | 0 [| 4973] 1269 | 682 8 ER n 
Gauss Blur Total Area 180 58 19.03 128.9 6.53 4.63 21 0.70 
Original Image Residential Area 162 64 14.40 | 120.8 5.83 ; : 
Gauss Blur Total Area 100 38 12.04 57.4 6.27 730 26.5 205 
Deviation Image Residential Area 95 74 4.74 83.9 4.22 
As Table 2 shown, the difference values of statistic data of the 2.4 Residential Area Extraction from Texture Feature 
gauss blur images lessen while the grey range is half of the 
original image (the grey value range of original image is 0 to 
255 while the grey value range of gauss blur deviation image is 
only 74 to 95), thus the absolute difference of deviation 
increases too, which means the gauss blur can further improve 
the effect of the threshold selection. The most important point is 
the high increasing of the average difference especially the 
gauss blurring result of the deviation texture image (difference 
of average is 26.5, the variation of the average between the 
original image and gauss blur deviation image is 28.3 which is 
large than the difference of the deviation and average in the 
texture image, Section 2.2). In this step, the average can be used 
as the threshold feature for the segmentation of the gauss blur 
image to obtain best extraction result of the residential area. 
  
Images 
With the self-adaptive threshold, the coarse segmentation image 
can be obtained. While the outer roads always connect to the 
residential area and have similar feature to the roads inside the 
residential area. Thus some outer roads are included in the 
residential area. Skeleton processing is proposed to eliminate 
the road from the residential area. The outer roads have low 
width of skeleton, thus we can identify the outer roads with the 
width of skeleton to obtain high precision border of the 
residential area. Figure 4 is the whole processing flow chart of 
the residential area semi-automatic extraction in the high- 
resolution image 
  
| Select Seed | [Select Processing Window | 
Deviation Calculation 
N 
  
Deviation Image 
     
   
    
Gauss Blurring 
Gauss Blur Image 
Iterate to get threshold 
     
    
      
  
    
  
    
Including Road Neighbor 
to Residential Area? 
Y 
  
| Obtain skeleton of the Region | 
  
Vertical distance of pixel to skeleton 
is less than the given value? 
  
V Y 
  
  
  
  
[pixels vertical to skeleton is non-residential pixels | 
residential pixels 
  
[Threshold to Gauss blur deviation image | 
  
Region Growing 
| Obtain Residential Arca | 
[T RS 
    
  
  
  
  
  
Pixels vertical to skeleton is 
  
  
| 
A 
    
Segmentation End 
  
s Extraction Board of Residential Ares | 
  
Fig 4 Processing flow chart of the residential area semi-automatic extraction 
1124