cond at state of 
| grown manifolds 
ion and area. 
and 18?37"N and 
id cover an area of 
population of Pune 
o 1991 it grew by 
wth increased to 
he recent census 
22.6% (Census of 
ining Organization 
inistry of urban 
largest number of 
2009-10 had stated 
y were growing, at 
vould be living in 
tructures made of 
1 as wood used for 
1s, galvanised iron 
the most basic of 
lum population is 
ial resolution) was 
‚The image shows 
ical urban features 
Cloud free Pan 
the year 2006 was 
s. 
o used as thematic 
Census data and 
city Municipal 
this study. 
ms in OOA is to 
n of an image into 
s) and the criteria 
omogeneity within 
ong neighbouring 
as done by using 
jective choice by 
pol (Drágut et al., 
itives’, which lead 
urther refinement 
all image layers as 
    
well as the thematic layer of road and water layer to create 
meaningful segments (Teo &Chen, 2004). At the analysis 
level, a classification of the generated image objects follows 
and ends 
in an iterative process of a knowledge-based object 
enhancement and (re-) classification. 
As the main aim was the image-based detection and 
classification of slums (Figure 2), the first step was to 
understand the human interpretation process behind visual 
identification of slums and mimic these in Definiens 
Developer. The introduction of knowledge in the classification 
process would help to depose the difficulties of information 
extraction from very high spatial resolution images. In general, 
classification means to assign a number of objects to a certain 
class according to the class’ description. Thereby class 
description is performed by describing typical properties or 
conditions the desired classes have. The objects then become 
assigned (classified) according to fulfilling or not these 
properties/conditions. 
  
   
        
Multi rasolution 
segmentation 
  
   
M 
on Baittus | Ruleset | 
„development “ 
Shadows | V | 1 = ] 
i 
Esfning tke 
classification 
and Validation 
Accuracy 
Assessment 
Figure 2 Flow Chart showing the methodology 
  
  
  
  
  
  
3. RESULTS AND DISCUSSIONS 
3.1 Segmentation 
In order to receive image object primitives as basic processing 
units, the object oriented approach to image analysis needs a 
complete segmentation of an image. The term “segmentation” 
means: “an operation that creates new image objects or alters 
the morphology of existing image objects according to a given 
criteria” (Definiens, 2010). Image segmentation methods are 
split into two main domains: knowledge driven methods (top- 
down) vs. data driven methods (bottom-up). In comparison, the 
basic differences between both approaches are: top-down 
methods usually lead to local results because they just mark 
pixels or regions which meet the model description. In 
contrast, bottom-up methods perform a segmentation of the 
complete image. They are more grouping pixels to spatial 
clusters, which meet certain criteria of homogeneity and 
heterogeneity (Definiens, 2010). In the present case, both these 
approaches were used in Image segmentation. ESP tool was 
used to get the scale parameter as an objective choice instead 
of trial and error and with scale parameter of 40, 0.5 as 
shape/colour ratio and 0.5 for compact and texture ratio the 
multi-resolution segmentation was done for the whole image. 
All three bands (BGR) of quick bird data and the thematic 
layers (roads and water bodies) were used for image 
segmentation. The resulted objects were good enough to go for 
classification. 
3.2 Classification 
Feature recognition is an essential part of object-based image 
analysis. A comprehensive feature extraction methodology is 
the precondition for successful work with image objects. Given 
the large number of possible features for object description, it 
is necessary to identify the characteristic, significant features 
for object-classes of interest ( Nussbauma.S., et.a1,2008) 
In the present study in order to identify the slum areas, the 
following factors were considered for classification: 
Y Small sized structures with high density 
v Tone difference in the slum areas 
v' Irregular internal street pattern 
Y' Less/no vegetation/green in slum areas 
Y. Areas of wastelands, such as on banks of Rivers or 
Canal, along railway line and road margin were taken 
as association for identification. 
In Pune, the existing slum locations (Table 2, Figure 3) have 
formed the basis of identification of these factors. 
  
  
  
  
  
  
  
  
Location Number of 
Hutments 
River banks 5142 
Along Small streams 
and canals 19670 
Hill and Hill slopes 11604 
Roadsides and 
places Meant for 3584 
Public use 
  
Source: Environmental Status Report, 2006-2007 
Table 2. Location of slums in Pune city 
  
Figure 3. Location of Slums near canal and along the hill 
slopes 
The image classification in e Cognition is based on user- 
defined fuzzy class descriptions of spectral, spatial and 
contextual features. The classification started with assigning 
image objects to roads and water bodies. In the absence of NIR 
band, using thematic layer to classify the image objects as 
water bodies helped to reduce the false positives as well as 
further refining the rule set. Since majority of the slums are 
located near the water bodies and along the road, railway line 
sides, classifying these objects in turn helped to detect the 
location of slums. 
Classification of non-built-up area such as Shadows and 
vegetation was carried out by using brightness values and