The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2008 
1137 
4. RESULTS 
4.1 PRISM Image quality 
PRISM images showed some radiometric artifacts that can be 
categorized in two different groups: black reference calibration 
between detectors,*and image compression artifacts. 
The first category of radiometric artifact visible on the PRISM 
image is induced by an incorrect calibration of each CCD 
sensors, resulting in column stripes with different inconsistent 
illumination. It has been reported by some scientist with early 
PRISM images (Wolff K. and al. 2007). However, it looks like 
JAXA is now providing better images as this artifact is no more 
visible in images currently available. 
Second category of radiometric artifact seems to be induced by 
on-board image compression which uses a loss compression 
scheme (like JPEG compression scheme). It is impossible to 
correct this artifact through some post-processing algorithms as 
some information in the initial image has been lost. This artifact 
can have a bad influence on the stereo-matching algorithm as it 
will introduce some false patterns that may confuse the 
matching process, as shown in next figure. 
Figure 4: Block artifacts in zoomed window 
A side-effect of this compression artifact is also some geometric 
distorsion in the image. It can be seen along some image blocks 
(8 pixels width) on a complete stripe of pixels. As illustrated by 
the next figure, this artifact seems to induce some local 
geometric image distortion: blocks of images are shifted by few 
fractions of pixels. As previous artifact, this will also have a 
bad influence in the matching process. 
4.2 Geometric quality assessment 
By computing individual Digital Elevation Model between all 
pairs of images (backward, nadir and forward) and analyzing 
them, it was possible to highlight some repeating undulations in 
the resulting elevation images as illustrated by the following 
figure. 
Figure 6: Undulation observed in resulting elevation image 
These undulations are probably resulting from vibrations of the 
satellite platform that are not well measured and not available in 
the ancillary geometric information provided with LevellBl 
data. Those vibrations have an impact on the resulting elevation 
image of 3 meters (with a frequency of 8 Hertz) and 1 meter 
(with a frequency of 93 Hertz). 
4.3 Digital Elevation Model extraction 
In order to have a consistent comparison between Digital 
Elevation Model from IKONOS and ALOS dataset, the bundle 
adjustment has been done with all data together. This prevents 
from introducing position bias error between the resulting 
elevation images. 
In order to attenuate effects of undulation illustrated in this 
article, and also false matching due to the radiometric artifact, it 
seems important to merge individual elevation measures with 
an output posting around 10 meters (from initial GSD of 2.5 
meters). Resulting merged elevation image will provide better 
robust and accurate elevation measures. Next figure show the 
resulting Digital Elevation Model for those datasets. 
Figure 5: Geometric distortion inside highlighted rectangle Figure 7; ALO s PRISM DEM and overlapping IKONOS DEM