; JPEG 
ital, Distortion 
| Without effecting the 
ested for compressing 
are compared to the 
vents. It is found that 
r visual applications. 
000 bytes of three TM 
visual quality. Beyond 
plications where other 
s, however, that JPEG 
1 visual quality of the 
ralized for all images. 
ct on the amount and 
‘the processed data is 
ittle but sufficient and 
the image processing 
^ the possibility of 
and many other related 
e solved. 
paper is to study the 
JPEG technique for 
sensed data. This 
be useful in reducing 
ll images for visual 
. A study was made on 
( of smooth distinctive 
vs that a 10% reduction 
grading the visual or 
1mi, and Sarjakoski, 
interest to test the 
remote sensing images 
ons. In this study, a 
tion is used where the 
. Vienna 1996 
  
effect of compression on subsequant processes 
such as image classification is tested. The paper 
is organized in the following manner. In the 
next section, the JPEG concept is presented and 
evaluated. Image classification is introduced in 
section three to facilitate evaluating the JPEG 
effect on the compressed TM images. The 
experiment and analysis are evaluated in section 
four, and the conclusion is made in section five. 
2. JPEG CONCEPT 
JPEG is an international standard for achieving 
image compression to reduce the amount of 
stored data and the period of transmission of 
such data. JPEG was found useful in 
compressing different types of images especially 
those of terrestrial successive frames (Langdon, 
et. al 1992) by taking advantage of the data 
redundancy in the coding process (Pennebaker 
and Michell, 1988). The overall scheme is 
basically transforming 8*8 pixels from space 
domain to frequency domain. There are two 
main processes performed by the technique, 
namely encoding and decoding as shown in 
Figure 1. 
QUANTIZATION 
INPUT COMPRESSED 
IMAGE 
DET] ENCODING Terras 
IMAGE 
ENTRO 
«—|»cr| DECODING 
CONSTRACTED COMPRESSED 
IMAGE IMAGE 
  
  
  
  
  
Figure 1. Encoding and Decoding for Image 
Compression 
In the encoding process, the raw data passes 
through the discrete cosine transform (DCT) 
function to transform it to a domain in which it 
can be more efficiently encoded. The DCT 
follows the following mathematical model. 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996 
M 
1 
7 
X(u,v) —K(u) K(v) xz G3) 
: hà (1) 
(2i*1)ux {27+1) vn 
Copal rr Cose 
where i, j, u, v e [0, 7], x(i,j) — (i, j)* element 
in an 8x8 block, X(u, v) — (u, v)? coefficient in 
an 8x8 DCT coefficient matrix, and K(u) — 
14/2 for u = Oand 1 foru z 0: 
Then, the data is scaled down to 
lower-precision demanding fewer bits, a proces 
called quantization. It employs the following 
equation in which C(u,v) is an integer and 
Q(u,v) is a suitable number. 
(2) 
The resulting data is then coded using Huffmn 
representation. Such a process leads to a 
compressed image. 
The decompressed image is subjected to the 
inverse of the DCT function and the 
quantization processes. The mathematical model 
for inverse DCT (IDCT) is as follows: 
7 7 
x(i,j)-2) ) ku)ktv)xtu,v), 
  
u=o v=o (3) 
OS (2i*1) ux os (23*1) vm 
16 16 
Dequantization is the opposite of quantization 
presented by the following equation: 
Xiu. vv) = Clu, vV) Qu v) uu)