stanbul 2004 
mage can be 
] the quality 
le. A higher 
ictor gives a 
. Therefore, 
'aling the Q- 
ssion scheme 
lucing image 
ferent JPEG 
d Images 
arious image 
Grgic ef al., 
2002, 2003) 
pression for 
ts of lossy 
the accuracy 
investigated 
iccuracy of 
| image after 
03) reported 
determine a 
image. 
ossy image 
no texture 
ot be applied 
| to carry out 
compression 
and texture 
le of 1:7000 
um on the 
Each color 
1 into 8 bits 
96 lines was 
ze. Figure 2 
)6 lines 
  
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004 
  
Figure 3. Block C2 
Figure 4. Block H3 
  
  
  
  
  
  
  
  
  
  
  
  
Image size 512 pixels x 512 lines Image size 512 pixels x 512 lines 
Gray level features [R] [G] [B] Gray level features [R] [G] [B] 
e mean 170.0 156.5 152.0 e mean 88.6 96.2 115.9 
e standard deviation 65.1 52.1 38.4 e standard deviation 55.6 54.5 45.2 
Color features [H] [5] [V] Color features [H] [S] rv] 
e mean 94.9 48.5 177.9 e mean 145.1 76.9 118.2 
e standard deviation 80.4 21.4 I 565 e standard deviation 41.1 28.4 49.2 
GLDV [R] [G] [B] GLDV [R] [G] [B] 
e contrast 22.4 21.5 17.1 e contrast 13.8 14.6 12.3 
« angular second moment | — 0.188 0.125 0.136 e angular second moment {| 0.216 0.185 0.176 
e entropy 4.24 4.43 4.22 e entropy 3.60 3.82 3.77 
* mean 14.7 15.6 12.8 e mean 7.3 8.7 8.1 
_ Fourier power spectrum [R] [G] [B] Fourier power spectrum [R] IG] [B] 
e mean spatial frequency 22.3 29.2 31.8 e mean spatial frequency 14.6 15.8 15.4 
  
Table 2. Statistics of Block C2 
shows the test image used in the experiment. Statistics of the 
test image are shown in Table 1. 
We divided the test image into 8 by 8 blocks, so that each block 
Was a square image of 512 pixels by 512 lines. 64 blocks were 
investigated in the same way as the whole test image. Block C2 
located in the upper left of the test image was the least 
compressible block, in other words, compressed image data 
volume of Block C2 was the largest of all 64 blocks. On the 
other hand, Block H3 located in the lower left of the test image 
Was the most compressible block. Locations of both Blocks C2 
and H3 are shown in Figure 2. Figure 3 and Figure 4 show 
Blocks C2 and H3 respectively. Table 2 and Table 3 show 
Statistics of Blocks C2 and H3 respectively as well. The texture 
of Block C2 is finer than that of Block H3. 
3.2 Lossy Compression 
Four compression types shown in Table 4 were investigated. 
Three Types S444, S411 and RGB adopted lossy JPEG 
compression technique, and the Type J2K adopted lossy JPEG 
2000 compression technique. 
649 
Table 3. Statistics of Block H3 
We used Independent JPEG Group's free JPEG software for 
JPEG compression. The quality setting of the IJG's software 
scales quantization tables to adjust image quality. The IJG 
quality setting, which is opposite to the ordinary Q-factor, is 0 
(worst) to 100 (best); default is 75 (Independent JPEG Group, 
1999). The IJG quality setting lets one trade off compressed 
image data volume against image quality of the reconstructed 
image: the higher the IJG quality setting, the larger the JPEG 
file, and the closer the reconstructed image will be to the source 
image. 
The test image was compressed at 21 various levels using IJG 
quality setting of every 5 from 100 to 0. In Types S444 and 
S411 color space conversion was performed. While Type 444 
adopted the 4:4:4 sampling (no downsampling) Type S411 
  
  
Type Technique Colorspace Downsamplig 
$444 JPEG YCbCr 4:4:4 
S411 JPEG YCbCr 4:1:1 
RGB JPEG RGB 4:4:4 
J2K JPEG 2000 
  
  
Table 4. Compression types