The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008
322
Figure 3. Influence of punctual constraints on final radiometry
3.3 Test case over Tasmania
The process has been tested on Tasmania Island, southeast of
Australia. Due to oceanic climate, input data is very
heterogeneous. Eleven very cloudy SPOT5-HRS segments
taken from 2004 to 2007 in winter (July) as in summer (January)
were required to cover nearly 70000 km 2 . With one point every
kilometre, an initial water mask and two iterations, results are
promising. Degree 1 polynomials make globally well-adjusted
images (Fig. 4). RMS of radiometric differences on overlapping
area drops from 62 to 14. And consequently, seams between
images are less visible if not invisible (Fig. 5).
Figure 4. Initial and final radiometries - Tasmania
07/06/2004
07/06/2004
i
13/04/2007
13/04/2007
Figure 5. initial and final limit between 2 images - Tasmania
For each image, a cloud mask is created automatically from
invalid grid nodes (Fig. 6). These masks have proved to be very
useful for mosaicking step.
Figure 6. Image and cloud mask - Tasmania
4. APPLICATION TO DEM TILTING
Reference3D DEM is obtained by merging overlapping DEMs,
calculated by automatic image matching of SPOT5-HRS stereo
pairs. But residual errors in “geometric” block adjustment result
in small altimetric discontinuities between DEMs so tilting
them is necessary before merging step. Radiometric block
adjustment methodology can be adapted easily for DEM tilting.
Image radiometry is simply replaced by DEM altitude and
polynomial model applied to altitude is as follows (eq.7):
FinalAlt(col, row)= InitialAlt(col, row)+Q(col, row) (7)
where:
• InitialAlti is initial DEM altitude for pixel (col,row)
• FinalAlti is calculated altitude for pixel (col,row)
• Q is a polynomial limited to degree 1.
Multiplier polynomial P, present for radiometric block
adjustment, is not used for DEM tilting. Addition to initial
altitude is the only permitted operation, multiplication is
forbidden.
5. PERSPECTIVES
Radiometric block-adjustment strategy described in this article
has been validated on medium-size datasets. Previously manual
tasks like radiometric adjustment or DEM tilting are now
mostly automatic and give good results. But, many studies
could be conducted to improve this algorithm, in particular
concerning system constraints on standard deviation invariance
to have a better control of image dynamic. And practically, this
functionality has yet to be transferred to Reference3D mass
production, typically from Tasmania dataset to Australia dataset.
REFERENCES
Ackerman, S.A., Strabala, K. I., Menzel, P. W.P., Frey, R.A.,
Moeller, C.C. and Gumley, L.E., 1998: Discriminating clear
sky from clouds with MODIS, Journal of Geophysical
Research, 103(D24):32,141-32,157.
Caprioli, M., Figorito, B. and Tarantino, E., 2006. Radiometric
Normalization of Landsat Etm+ Data for Multitemporal
Analysis, ISPRS Proceedings “From pixels to processes”,
Enschede (NL).