International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
adopted the 4:1:1 sampling. In Type RGB the source image
was separated into three color component (Red, Green and
Blue) images, and each color component image was
compressed and decompressed respectively. Evaluation of
quality of the reconstructed image was executed after
combining three decompressed color component images.
We used LuraWave SmartCompress 3.0 by Algo Vision
LuraTech GmbH for JPEG 2000 compression. We prepared
JPEG 2000 compressed image data of the same size of JPEG
compressed image data of Type S444.
As for Blocks C2 and H3, only Type S444 was investigated.
3.3 Measures of Image Quality of Reconstructed Images
The following numerical image quality measures were
examined.
(A) Gray level features
a) PSNRs of RGB (Red, Green, Blue) components
(B) Color features
a) PSNRs of HSV (Hue, Saturation, Value) components
Each component of HSV color space has 256 levels (8
bits).
(C) Texture features
a) Standard deviations of RGB components
b) Grey level difference vector (GLDV) measures:
150
S411
e —9-- Sá|l /
8 100 —— Rod /
© — Blue / {it Blue
e # 1 Green
E. 50 / 4. JAA Red
= pra J
ue etd I i À
IE RR x pem
0 eins T |
100 80 60 40 20 0
1JG quality setting
Figure 5. Compression ratio vs. JG quality setting
(IJG quality setting: 100 - 0)
20 , S411
H3
os — SH ] l a
e 15 77 RGB oes
S m. CÓ E. yt D o
= eus jou e Do
2 P L^ € CE
e p rr ed RGB
o ; n ane
o s] v crat
2 d tem
(7 Wu pert
X
o !
100 90 80 70 60 50
[JG quality setting
Figure 6. Compression ratio vs. IJG quality setting
(IJG quality setting: 100 - 50)
Contrast, Angular second moment, Entropy, Mean
c) Mean spatial frequency of Fourier power spectrum
An image of finer texture features has higher values of
texture measures except GLDV Angular second moment
and a lower value of GLDV Angular second moment.
3.4 Results
Since the effect of changing compression ratio on PSNR of each
44 —— S444T ]
E
= 40
©
=
6
rx
Z - 36
N
a
32
Compression ratio
Figure 7. PSNR of Hue vs. compression ratio
50
+ RGB
2 erc Si
= —— RGD
eg 45 m
8 memes
a — M3
S
N
©
cz 40
Z
un
e
35
0 4 8 12 16
Compression ratio
Figure 8. PSNR of Saturation vs. compression ratio
56
a 32 -—-- $411
© :
= -—— RGB
S 48 ei
3
> —— H3
©
> 44
Zz
N
a
40
36 |
0 4 8 12 16
Compression ratio
Figure 9. PSNR of Value vs. compression ratio
650
In
Cc
R
Standard deviation
GI.DV Contrast