The International Archives of the Photogrammetry. Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008
(a) Singapore (8192 x 8192) (b) Switzerland (1000 x 1000)
Figure 1: Quickbird data used for simulations, original Pan and Xs images are orthorecti fled and fusioned
4 JPEG 2000 OPTIONS AND SIMULATIONS
4.1 Intercomponent transform
One of the most critical options for efficient use of the JPEG 2000
standard on high resolution satellite image resides in the mul
ticomponent transform. The part 1 of the standard only makes
provision for the three bands color transform. In part 2, it be
comes possible to specify an ad-hoc transform. Using a specific
transform with fixed coefficients computed on a set of images
(Thiebaut et al., 2007).
Spectral bands of high resolution satellite images present a strong
correlation. This is particularly the case after the PAN-sharpening
step. Thus, it is important to decorrelate these images before
compressing them. After the intercomponent transform, images
B' 0 , B[, B' 2 and B 3 are compressed with JPEG 2000 and the rate
allocation is common to all the transformed bands. Three differ
ent situations of reasonable complexity are studied:
• No intercomponent transform to keep the compatibility with
part I of JPEG 2000 standard.
Bo
B 3
A YCbCr transform on the first three band and no transform
for the fourth one.
B 0
B 3
1.772
-0.344136
-0.7141 3 6
1.402
A transformation by an average Karhunen-Loeve (KLT) ob
tain on a big image data set with similar spectral character
istics (sensor dependant) (Thiebaut et al., 2007):
b 0
-0.59952748
-0.2377866
0.75526749
-0.11659861
0.32418259 -0.3069099
0.58451217 -0.38087386
0.52300888 -0.23381439
0.52887922 0.84028111
0.66428901
-0.67582405
0.31837578
0.02488911
(3)
Fig. 2(a) and 2(b) show that using the part 2 extension enables a
gain of 5 to 8 dB PSNR.
4.2 Vector data
Using the Geographic Markup Language (GML) standard (GML
in JPEG 2000 for Geographic Imagery (GMLJP2) Encoding Spec
ification, 2006), JPEG 2000 is able to handle complex metadata
directly in the compressed stream. This is an advantage as com
plex product can be presented in a simple way.
The GML standard defined by the Open Geospatial Consortium
(OGC) can modelize, carry and save geographical data in the
XML format. This standard can describe:
• geographical objects
• projection systems
• geometry
• topology
• time
• measurement units
• attributes of geographical objects.
Thus, the GML standard is well adapted to carry auxiliary data
with a satellite image. Information can be: clouds mask, road
extraction result, segmentation result (eventually obtained by the
data provider with external information), region of interest de
scriptions. ..
4.3 Data organization
In JPEG 2000, several possibilities are available to structure the
data organization. Several progression orders are available with
JPEG 2000. They are denoted with the letters LRCP: L being the
quality Layer, R the Resolution, C the Component and P the Par
tition of the image. Order of the letter indicates the organization
of the different property progression. The default order LRCP is
illustrated on Fig. 3.
The five available progression orders are:
1284