COMPARISON OF OBJECT ORIENTED IMAGE ANALYSIS AND MANUAL 
DIGITIZING FOR FEATURE EXTRACTION 
H. Sahin* *, H. Topan *, S. Karakis ", A.M. Marangoz* 
* ZKU, Engineering Faculty, 67100 Zonguldak, Turkey - (sahin, htopan, karakis, marangoz)@jeodezi.karaelmas.edu.tr 
PS YF 
KEY WORDS: High Resolution, Mapping, Analysis, Digitisation, Extraction, Comparison, Orthoimage 
ABSTRACT: 
The developments in the remote sensing technology have provided the use of high-resolution images for different purposes if 
possible. These images can be used for a study such as town planning where high resolution and information content are required. In 
this study, high resolution panchromatic KVR-1000 image has been employed for extraction of man-made structures in a 
metropolitan city area. The test area is a part of Zonguldak (Turkey) city. First, boundaries of buildings and road’s center lines have 
been digitized manually. Additionally, the object oriented classification process has been implemented for the same area. In this 
manner, the results from manual digitizing and large-scale maps produced by photogrammetrical techniques have been compared 
and the success of manual digitizing has been verified. The large-scale maps have been taken as the base criteria in the comparison. 
The second analysis deals with tests using object oriented classification. Both methods include some disadvantages. Operator could 
experience some problems during manual digitizing process. The object oriented analysis is an alternative tool which uses grey 
values of objects in the process. Both methods have been analyzed for the orderly and disordered zones constituted by the buildings 
and the independent houses, respectively. 
1. INTRODUCTION 
One of the main objectives of geoinformatics engineering is to 
collect data and analyze, represent the products of these data 
with diverse means. Today, the reliable production of 
information and its rapid serve to the user community is an 
important task. The reliability and rapidity aspects of 
information provision have accelerated the progress of 
technology. As a result of this stormy progress data acquisition 
from space has been an operational concern. Data acquired 
from space can be used in different disciplines such as 
geoinformatics, forestry, agriculture and etc. 
Extracting both geometric and semantic information from space 
images has been the main concern since the early phase of 
remote sensing. Qualitative analysis of images does not help the 
user to deduce required information. Abundance of information 
in both photographs and space images leads user to digitize 
only the interested objects. Digitized objects help user to extract 
information on only focused features. Map digitizing started 
with the invention of digitizing tablets. The graphical map to be 
digitized is laid on the digitizer table and coordinate values of 
discrete points are stored in a computer using a cursor. With the 
emergence of scanners on the market, graphical maps are 
scanned at equal intervals throughout the whole image and 
scanned data are stored in raster file format. The next step has 
consisted of vectorizing the raster image. Several methods are 
developed to vectorize raster images. One of those methods is 
called on-screen digitizing as a manual method, and the other 
one is called object oriented image analysis as an automatic 
method. The basic purpose of developing these methods aims at 
reducing operator’s interactivity with the computer and thus 
speeding up the digitizing process. 
  
* Corresponding author. 
114 
Today, space images can be used for data acquisition purposes. 
Many vendor companies provide a wide range of images for 
different users. One.of the high resolution space imagery is the 
Russian KVR-1000 system. KVR-1000 images are for example 
used to isolate illegal buildings in a forest environment in 
Greece (Karathanassi et al., 2003). Kostka (2002) uses KVR- 
1000 imagery together with other high resolution images in 
order to make inferences about climatic studies, transportation 
routes, water resources, conservation areas, and relicts of 
human land-use. Another study uses KVR-1000 images to 
monitor refugee camps in south-east Asia (Bjorgo, 2000). 
All above studies deal primarily with semantic relating of 
spatial objects. Geometric accuracy potential of KVR-1000 
images is not well researched. Therefore a geometric 
assessment should be made within the context of this study. For 
this goal objects in KVR-1000 images are digitized using 
manual and automatic methods. Both obtained results are 
compared to 1:1000 photogrammetrically produced digital line 
maps. 
2. METHODS 
Digitizing is a way of conversion of information from 
analogously produced graphical maps to machine readable 
vector or raster formats. Many methods are used for the 
vectorizing process. Two of these methods are adopted in this 
study, These methods are manual on-screen digitizing and 
object oriented approach. 
In 
n wed FF CN ^ uai 
e—— 0 tm 07] M ih 
C md CO NN ln AN 0 Vv 08 8 (0 CH 8 
—