green ratio. Shadow of the trees and buildings could be 
detected using the low values of brightness and mean 
difference to scene. Shadows were used as a proxy for height 
information in the absence of DSM. The high rise buildings in 
planned areas exhibited clear shadows (NB1) compared to 
narrow or negligible shadows of slum buildings. Vegetation 
(NB2) was classified using the customized arithmetic ratio 
values (green ratio i.e., green / blue + green + red). These 
classes were useful in defining contextual relationships for 
characterization of built-up. 
Classification of non-built-up area is followed by classification 
of built-up area. In high-resolution satellite images, a very 
salient feature of urban scenes is that they are highly textured. 
It is therefore adequate to use the texture features of the image 
as a measure to classify the image (Shan Yu and Berthod, M. 
and Giraudon, G, 1999). Texture based measures such as Grey 
Level Co-occurrence Matrix (GLCM) Entropy (quick8/11) of 
blue and red bands was used to classify built-up. GLCM is a 
tabulation of how often different combinations of pixel 
brightness values (grey levels) occur in an image (Definiens, 
2009). The quick (8/11) is a performance optimization version 
and works only on data with a bit depth of 8 or 11bits. The 
entropy tells how the elements are distributed in a given space. 
If the elements of GLCM are distributed equally, then the 
value for entropy is high. Hence in the present case, GLCM 
entropy value of above 4.1 in the blue band was used to 
classify the image objects into built-up (Figure 4). The 
classified built-up objects are again refined by using the 
entropy value less than 3.8 in band 3. 
    
entaticn 
zzz 40 [shape:05 compet.:0,5] creating New Level” 
» class 
iA with Id: Thematic Layer! = 1 at Mew Level: Water 
Jis with Ic: Thematic Layer 1 = 2 at New Level: Road 
i ASF Water at New Levek grow into all 
ii. = Non-Built area 
i Y& unclassified viith Brightness « 310 at. Neve Level: NBI 
- 2%. unclassified with Mean diff. tc scene Layer3 < -34 at New Leveh NBI 
ME, NBI, unclassified with Arithmetic Feature 1» 042 at New Level: NB2 
with Mean Layer 2 > 500 at New Levek unclassified 
    
LE, unclassified «ith GLCM Entregy (quick 8511) Layer {0% > 41 at New Level: Builtup 
ME, Builtup with GLCM Entropy (quick 84113 Layer 3 10°} < 38 at New Levek unclassified 
Nis unclassified with Mean diff. to scene Layer 3 > 40 and Rel. border to unclassified < 0.79 at Ness Level Builtup 
Figure 4. Rule set for classification 
3.3 Separation of slums and non-slums 
Non Slum areas display different characteristics than slum 
areas. À combination of spectral, morphological and contextual 
information was used to classify non-slum (formal) areas. The 
association with vegetation and shadow were important 
indicators for separation. A sequence of steps involving 
brightness, layer means, texture and distance to features was 
used to further refine the classification. The mean layer 
intensity values of an image object in blue and green bands 
were used to separate the non-slum built up from other built up 
area. The formal built-up objects are having higher value in 
these two layers than other built up objects. Similarly the 
formal built-up objects are brighter than other built up due to 
the construction materials; so, brightness feature was used to 
separate formal from informal (Slums). 
When classifying a satellite image, the class assigned to an 
object depends not only on the spectral feature of the object 
itself, but also on the spectral feature of its neighbours. In this 
context, spectral characteristics of Image such as GLCM 
Texture measures also used to separate the formal from 
Informal. GLCM Contrast of band 3 was used for this purpose. 
  
   
   
    
  
   
   
     
    
   
   
    
   
  
  
  
  
  
  
  
    
   
  
  
  
  
   
   
   
    
   
    
    
  
  
  
  
  
  
  
  
     
   
    
   
  
    
    
    
  
The contextual information was used to separate non-slum 
built-up objects from Slum (informal) objects. Hence the 
feature Rel. Border was used to determine the relative border 
length of formal built-up object shares with the NB2 and NBI 
(Green vegetation and Shadow in Figure 5) image objects. It 
describes the ratio of the shared border length of an image 
object (with a neighboring image object assigned to a defined 
class) to the total border length. If the relative border of an 
image object to image objects of a certain class is 1, the image 
object is totally embedded in them. If the relative border is 0.5 
then the image object is surrounded by half of its border. The 
relative border length can only be expressed in pixels. The 
relative border value of above — 0.399 was used to separate 
formal built-up from Informal built-up objects. 
  
  
Figure 5. — Proximity of vegetation and shadow in built-up 
which differentiate between slum and non-slum areas 
When contextual information is used at the object level, it can 
be used in association with common-sense knowledge to obtain 
a coherent interpretation of the whole scene or to aid in 
detecting less obvious objects. Mostly slums are tends to be 
located near vulnerable areas like close to river and 
transportation lines (Figure 3 & 6). Therefore Distance to 
water (River, canal) and roads features were also used to 
separate and refine the formal built-up areas. 
  
     
Figure 7. Slum classification 
After separating the formal areas, the remaining built-up was 
classified as slums. The built-up other than formal was first 
merged, so that geometry of the objects can be better 
unders 
themse 
areas | 
identif 
positiv 
elimin 
positiv 
exemp 
2008). 
areas, 
shape 
the ob 
by non 
this, 1 
param 
non-sli 
was cl 
other r 
  
Fig 
34 A 
The es 
include 
determ 
results 
realitie 
NGO's 
status 1