The International Archives of the Photogrammetry. Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008
(a) (b)
Figure 2: Performances comparisons according to the intercomponent transform. Using a precomputed KLT on huge set of similar data
enables a bitrate gain of a factor of 1.75 to 2.
• LRCP: progressive quality on the whole image;
• RLCP: progressive resolution on the whole image;
• RPCL: progressive resolution with a more localized access;
• PCRL: fast random accès on the image;
• CPRL: fast random access by component.
The five progression order are compared in term of speed access
and quicklook generation. On Fig. 4, the quicklook generation
time is detailled for different progression order. Differences be
tween progression remains small as the access was on hard drive
which are quite efficient for random access. Differences arë ex
pected to be much bigger when data are accessed through a net
work.
4.5
Order
Figure 4: Quicklook generation time for a compressed Quickbird
image of size 27504 x 26636. The quicklook correspond to a
reduction in pixel number by 32 in each direction.
4.4 Visualization session
The interest of using compressed data is also studied in the situa
tion of normal data visualization. A typical visualization session
with ENVI is simulated and it appears that using directly com
pressed data enables a gain of 20 in time. Results are presented
in Fig. 5. The increase in computational complexity for the de
compression is more than balanced by the reduced data transfer
required from the disk. This advantage of compressed data would
be even stronger in the case of distant usage of data through a net
work.
Order
Figure 5: Simulation of navigation session in a JPEG 2000 im
age with random access. Despite the computational cost of the
decompression, it is still faster to work directly with compressed
data (disk access is reduced).
4.5 Tiling impact
Tiling is an important concept to ease memory constraints at the
compression and decompression steps. For such images, where
holding the whole image into memory is out of question, tiling
is mandatory. The impact of tiling on visual quality is explored
for the common rates. Three different tile sizes are compared on
Fig. 6(a) and 6(b): 8192 x 8192, 4096 x 4096 and 1024 x 1024.
Using smaller tiles enables a reduction in memory requirements
without a loss in image quality. A block effect could arise for low
bitrate but above 0.5 bpp these are not visible.
5 A NEW COMPRESSION PARADIGM
With these end-users requirements in mind, this study leads to
perceive the compression in a new way. The main purpose is not