International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B-YF. Istanbul 2004 
  
Comparison between multispectral — synthetic image 
  
  
  
Band 1 Band 2 Band 3 Band4 
Bias (ideal value: 0) 0.178 0.099 0.079 0.016 
Correlation coefficient 0.798 0.804 0.814 0.747 
(ideal value: 1) 
  
Standard deviation of the 
difference image 0.296 0.007 0.03 0.039 
(ideal value: 0) 
  
  
  
  
  
Comparison between NDVI multispectral — 
NDVI synthetic image 
  
Bias (ideal value: 0) 1.143 
  
Correlation coefficient 
  
3 0.931 
(ideal value: 1) 
Standard deviation of the 
difference image (ideal value: 2.611 
  
  
0) 
  
  
Table 5. The results of the statistical tests 
The spatial evaluation of the synthetic image is done in order to 
detect if the synthetic image maintains the spatial characteristics 
of the high-resolution image (panchromatic image). A high pass 
filter 7x7 is applied over the synthetic and orthorectified 
panchromatic image, with the view to exaggerate the linear 
characteristics. The correlation between the images is done, 
after the matching of the ‘high-pass’ synthetic to the ‘high-pass’ 
panchromatic image histograms. 
  
Correlation coefficient between ‘high-pass’ 
panchromatic — ‘high-pass’ synthetic image 
Synthetic BANDS 
I 2 3 4 
0.892 0.863 0.85 0.834 
  
  
  
  
  
  
  
  
  
| Panchromatic BAND 
Table 6. The results of the correlation between ‘high-pass’ 
panchromatic — ‘high-pass’ synthetic image 
5. CLASSIFICATION 
The relevant data derived from the satellite sensors are so many 
that the classical interpretation methods for the extraction of 
information are difficult and time-consuming. Classification is 
one of the most reliable methods of recording pixels in classes 
of land use, with multispectral classification. 
Classification is the process of defining the image pixels into 
various classes, representing the different types of land cover 
and use. Each class or category in one image represents a group 
of pixels that have the same spectral values. 
There are two kinds of classification: 1) the unsupervised 
classification, where the classes of pixels are determined 
according to their band values without the use of the external 
data. Once the pixel classes have been formed the land cover 
type of each class is identified with the help of pixels within the 
class whose land type has been determined by fieldwork. 2) the 
supervised classification, where the pixels with known land 
cover type, determined by fieldwork form the nuclei for the 
classification of the remaining pixels in one of the already 
identified classes, on the basis of their band values. 
(A.Dermanis: Remote sensing) 
142 
5.1 Preparation of the image 
In this project the supervised classification is applied on the 
synthetic image, which was produced with the process of 
fusion. It should be noted that the synthetic image was divided 
into two parts, the one of the urban and the other of the hilly 
area. Each part is classified separately. This is done, because 
otherwise there were problems in the exact definition of the 
classes, and so the correlation between pixels and classes was 
not right and accurate. 
5.2 The steps of the classification 
The first and most important step in the process of classification 
is the definition of all the classes, where the pixels are going to 
be inscribed. Consequently, the choice of the samples from 
specific parts of the image must be done carefully, for the right 
correlation between the pixels and the classes. 
For the first part of the synthetic image, the urban area, the 
following classes are chosen: 
] Trees 
2. Grass 
3. Concrete 
4. Buildings 
5. Ground 
6. Ring road 
For the second part of the synthetic image, the hilly area, the 
following classes are chosen: 
Forest 
Burnt forest 
Road 
m Pom 
By collecting the proper samples of cach class, the 
classification is realised according to the fuzzy logic, which 
contributes to better results. 
  
Classes 
E Forest 
Burnt forest 
Ring road 
Ground 
Concrete 
Sea 
B Grass 
EH Trees 
[] Buildings 
  
   
  
  
  
Figure 7. The classified image 
5.3 Evaluation of the classification's accuracy 
Comparing some specific pixels of the classified image and 
their corresponding reference pixels, which belong to a known 
class, succeeds the evaluation of the classification. 
In 
TAA T= 1 ~~ 1 
FT ST SSAITSSTESTÄASTFATZST 
  
AT