International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004 
different satellite image classification methods. For this 
purpose, landuse classification of the study area was conducted 
by different supervised image classification procedures and the 
results were compared with one another. © Landsat 7 ETM+ 
satellite image acquised in 11.16.2001 and the IDRISI 
Klimanjaro image processing and the GIS package were used in 
this study. There are different image processing and GIS 
softwares using all around the world. A lot of them have the 
similar properties and capabilities for use remote sensing 
purposes. The IDRISI Klimanjaro image processing and the 
GIS package is one of the most useful and economic software 
of these image processing packages. 
In this study, the Idrisi Klimanjaro was used for the different 
image classification chosen;  Parellelpiped, Maximum 
Likelihood, Minimum Distance to Means and Fisher Classifier 
(Linear Discriminant Analysis) classifiers were used to 
determine which classifier is more effective and useful for this 
study purpose. To test these classifiers, a land use application 
was made in the study area. In this context, CORINE method 
was used for land use classification. 7 land classes were 
selected. Artifical Surfaces (Urban Areas), Agricultural Areas, 
Forests and Olive Trees, Wetlands and Water bodies (sea, lake) 
are the selected land classes according to CORINE land use 
method (CORINE, 1995). Bare Land class was added to the 
selected classes. For this purpose, PCA (Principal Component 
Analysis) composite image which was composed by PC2, PC4 
and PC5 band combination was constituted because the each 
PC images reflects the most principle components on that band. 
Training sites have been digitized on screen and so a signature 
file to clasify the image have been made. After then, four image 
classifiers, Parellelpiped, Minimum Distance, Maximum 
Likelihood and Fisher, were applied to clasify the composite 
image respectively. 
2. METHODOLOGY. 
There is a consistent logic to all of the supervised classification 
routines in almost all image processing softwares, especially in 
IDRISI Klimanjaro, regardless of whether they are hard or soft 
classifiers (IDRISI Klimanjaro Guide, 2004). In addition, there 
is a basic sequence of operations that must be followed no 
matter which of the supervised classifiers is used. In this study 
the following sequence of operations were used. 
1. Defining of the Training Sites. 
2. Extraction of Signatures 
3. Classification of the Image. 
2.1 Defining of the Training Sites 
The first step in undertaking a supervised classification is to 
define the areas that will be used as training sites for each land 
cover class. This is usually done by using the on-screen 
digitized features. For this purpose, band is chosen with strong 
contrast (such as a near infrared band) or a color composite 
image for use in digitizing. In this study, a color composite 
image which was made with PC2, PC4 and PC5 images was 
used. Generally, one should aim to digitize enough pixels so 
that there are 10 times as many pixel for each training class as 
there are bands in the image to classify. This should be made 
with at least two or three training sites but, the more training 
site is selected, the better results can be gained. However, this 
procedure assures both the accuracy of classification and the 
true interpretation of the results. In this context, each land class 
have been represented with two and three training sites. 
2.2 Extracting of Signatures 
After the training site areas have been digitized, the next step is 
to create statistical characterizations of each information. These 
are called signatures in Idrisi (Idrisi Klimanjaro Guide, 2004). 
With this module, categorization of infomation which of each 
pixels is possible. In this step, the goal is to create a signal 
(SIG) file for every informational class.The SIG files contain a 
variety of information about the land cover classes they 
describe. Each SIG file also has a corresponding SPF file that 
contains the actual pixel values used to create the SIG file. It is 
used only by HISTO histogram (HISTO) in displaying 
histograms of signatures). These include the names of the 
image bands from which the statistical characterization was 
taken, the minimum and mean values on cach band, and the full 
variance /covariance matrix associated with that multispectral 
image band set for that class (IDRISI Klimanjaro Guide,2003). 
1092 
Inter 
2,3 ( 
The 
This 
desci 
The 
peda 
landi 
the N 
there 
MINI 
used 
stand 
that i. 
The 1 
there 
trainii 
classe 
24G 
In thi 
used | 
classif 
minim 
are use 
The | 
classif 
of the 
data (R 
Whilst 
simple 
is tha 
parelle 
classfie 
maxim 
unless { 
prior p 
account 
Finally, 
parellel 
(Richar 
The Mi 
This cla 
each bai