3. Istanbul 2004
results
O.K. (1990).
al Methods in
3° data® IEEE
g, Vol. 28: pp.
ssification of
ial Data Using
e ASPRS 1994
3.
W.B. (1995).
rtificial Neural
haracteristics",
sing, Vol. 6l.
assification of
:k-Propagation
1ce and remote
The Effect of
and-Use/Land-
ineering and
nage Analysis:
0-387-5480-8,
Multi-Source
University of
OBJECT-ORIENTED IMAGE ANALYSIS AND SEMANTIC NETWORK FOR
EXTRACTING THE ROADS AND BUILDINGS FROM IKONOS PAN-SHARPENED
IMAGES
A.M. Marangoz “ *, M. Oruc?, G. Buyuksalih *
*ZKU, Engineering Faculty, Department of Geodesy and Photogrammetry Engineering 67100 Zonguldak, Turkey -
aycanmarangoz@hotmail.com, murat_oruc@mynet.com, gbuyuksalih@yahoo.com
PS WG H1/4
KEY WORDS: Hierarchical, Contextual, Segmentation, Object-oriented, Classification, High resolution
ABSTRACT:
Traditional pixel-based approaches are based exclusively on the grey value of pixel itself. Thereby only the spectral information is
used for the classification. The situation becomes worse when extracting the certain features only. An object-oriented image analysis
is implemented in order to overcome the limitation mentioned above. The existing software, eCognition v3.0 allows the polygon
based classification process. It is based on fuzzy logic, allows the integration of a broad spectrum of different object features, such as
spectral values, shape and texture. This study demonstrated the extraction of buildings and roads from the high-resolution Ikonos
pan-sharpened image data by first dividing it into the segments and then classifying it using the spectral, spatial and contextual
information. The test site was agro-industrial area in the city of Zonguldak which has rolling topography along the Black Sea coast.
Land use classification results as well as the spatial information can be exported to GIS environment for evaluation purposes with
existing larger scale cadastral maps and other available ground truth materials.
1. INTRODUCTION
Classification relies on the pixel-based approaches is limited at
present. Typically, they have considerable difficulties dealing
with the rich information content of high-resolution data e.g.
Ikonos images, they produce inconsistent classification results
and they are far beyond the expectations in extracting the object
of interest. This situation brings meaningful operator
intervention to the implementation. Due to mentioned nature of
classical methods, new and object-oriented image analysis of
eCognition software can be used. Such algorithm requires one
or more image segmentations which should also be supported
by the additional information like contextual or textual to.make
the segments more appropriate for improve classifications.
Object-oriented approach takes the form, textures and spectral
information into account. Its classification phase starts with the
crucial initial step of grouping neighboring pixels into
meaningful areas, which can be handled in the later step of
classification. Such segmentation and topology generation must
be set according to the resolution and the scale of the expected
objects. By this method, not single pixels are classified but
homogenous image objects are extracted during a previous
segmentation step. This segmentation can be done in multiple
resolutions, thus allowing to differentiate several levels of
object categories. Automatic recognition and segmentation of
the common objects, eg. buildings and houses from high-
resolution images, eg. Ikonos and Quickbird was investigated
Some users with a certain degree of success (see, Hofmann,
2001a, b and c).
In this study, object-based classification of buildings and roads
in the Zonguldak testfield of Turkey has been realized by
eCognition v3.0 software. Classification procedure has been
* I. .
Corresponding author.
implemented using pan-sharpened Ikonos image of the interest
area. Such an image can be easily formed by the pan-
sharpening module of PCI Geomatica 9.1.1 system. Several
tests have been carried out to match with the successful
segmentation, then the classification by entering different
parameters to the used software. Authors, finally comments on
the pros and cons of the object-oriented based image analysis
with the detailed explanation of the obtained results.
2. EXPERIMENTAL AREA AND DATASETS
Zonguldak testfield is located in Western Black Sea region of
Turkey. It is famous with being one of the main coal mining
areas in the world. Although losing economical interest, there
are several coal mines still active in Zonguldak. Area has a
rolling topography, in some parts, with steep and rugged terrain.
While partly built city area is located alongside the sea coast,
there are some agricultural lands and forests in the inner part of
the region. The elevation ranges roughly up to 800m inside the
area covered by Ikonos imagery. Two Ikonos Geo PAN images
of this testfield were purchased from SI Eurasia which is the
regional affiliate of SI and located in the Turkish Capital,
Ankara. Important characteristics included in the metadata files
of these images are given as follows:
Ikonos Geo-product PAN
* * 1 € y ac
Characteristics images
Image I Image II
02/07/2002, | 02/10/2002
Date, Time 08:52 GMT | 08:59 GMT
