Where, X'(u, v) 
coefficients. 
is the dequantized DCT 
The JEPG concept is discussed by a number of 
researchers such as in (Wallace, 1992). This 
technique is applied to the original TM images, 
then the images are classified to make possible 
the processes of comparison and evaluation. 
3. IMAGE CLASSIFICATION 
Classification is, in general, the technique by 
which images can be easily analyzed and 
possibly interpreted. There are many techniques 
available for image classification (Congalton, 
1991 and Jensen, 1986). The classification 
techniques take advantage of the statistical 
characteristics of the image content and produce 
a thematic map containing a number of classes. 
Each class represents one feature of the scene. 
These visual and statistical characteristics of 
classification are utilized in this research where 
the effectiveness of JPEG is attested by applying 
the unsupervised isodata image classification 
technique to the compressed remotely sensed 
data. 
4. EXPERIMENT AND ANALYSIS OF 
RESULTS 
The input images are two 512 x 512 TM with 
three band each (2, 3, 4). For simplicity, the 
LAN and GIS images of experiment one will be 
abbreviated by E and that of experiement two by 
Ex in the text, tables and figures. Some times E 
and Ex are associated with numbers indicating 
the rates of compression. The original images 
(E.LAN and Ex.LAN) were classified prior to 
the compression, as shown in Figure 2 (E.GIS) 
and Figure 3 (Ex.GIS). Both E.GIS and Ex.GIS 
were considered to be free of error for the sake 
of comparsion. Then, the E.LAN and Ex.LAN 
were classified after being compressed at 
different levels of compression and several 
thematic GISmaps were obtained as also 
presented in Figure 2 (E8%.GIS, E10%.GIS, 
E12%.GIS) and Figure 3 (Ex9%.GIS, 
26 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996 
        
Ex12%.GIS, Ex14%.GIS). These sries of 
compressed GISmaps are compared visually and 
statistically with original E.GIS and Ex.GIS 
maps (the latter being assumed error-free). 
Notice that that two images of Figure 3 are 
omitted for simplicity. 
In Figure 2, for example, the E896 and E1046 
GISmaps compressed images are visually similar 
to the original E.GIS map. The E1296 
compressed image shows some differences when 
compared with the original classified image. The 
statistical analysis shows significant changes in 
classes such as 3, 5, 6, 7, and 9 as illustrated in 
Table 1. This table shows the number of pixels 
in each class for the original and the compressed 
image at different compression rates. The ideal 
case is to have no change in pixels’ values for 
all images. Table 2 presents the same 
information for experimeint Ex. Figure 4 and 
Figure 5 show the graphical difference in pixels 
between the uncompressed and the compressed 
images for selective classes from both 
experiements. The ideal shape for each figure is 
to have no deviation in the vertical axis, and to 
have only one horizontal line representing all 
images' pixels. This line should have zero slope 
and can be visualized as the horizontal 
compression ratio axis. 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
Class | E.GIS | E8% | E10% | E129 
1 4025 | 4204 |4482 | 4105 
2 5281 | 5698 | 5445 | 5298 
3 8696 | 9781 | 10841 | 8424 
4 5451 | 5417 | 4975 | 5704 
5 9540 | 9293 | 7921 | 10711 
6 10752 | 10108 | 9666 | 7651 
7 8077 | 6401 |7747 | 9098 
8 6871 | 7131 [6485 | 6197 
9 2214 | 3286 |3299 | 3564 
10 4599 | 5217 | 4675 | 4884 
Table 1. Number of Pixels of Original 
E.GISmap and Three Compressed GISmaps.