Yo) nd ES SEN, oM ims
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B-YF. Istanbul 2004
and is therefore very complex and time-consuming. Another
disadvantage is that for every sensor type different model and
different processing software has to be established. These
reasons motivated the replacement of the physical sensor
models by generalised models (Tao and Hu, 2001).
In 1989 Kratky suggested as a generalised model the use of
polynomial Rational Functions that.can support real time
implementations and provide accurate solutions. The Rational
Function model transforms the image pixel coordinates through
the ratios of polynomial of ground coordinates. In this project
Space Imaging Inc. provided the Rational Function coefficients
(RFCs) used in the orthorectification process.
3.2 The orthorectification processing
Firstly the following input data were used to perform the
orthorectification:
— The RFCs provided by Space Imaging Inc.
— The DTM that has a grid size of 25-m and covers all
the study area
— The GCPs obtained by GPS measurements and
additional GCPs obtained by orthophotos of scale 1:
5000 in order to cover evenly all the study area
The orthorectification with these data resulted in a large RMS
error of about 35 m. Aiming to minimise this error a
methodology was developed involving the following tasks:
— Synthesis from the two existing DTMs of a more
accurate DTM
- Refinement of the RFCs according to the ground
coordinates of the GCPs
— Orthorectification separately in three parts of the
image (the west, the east part of the city and the hilly
region)
3.2.1 Synthesis of DTM: In order to improve the accuracy
of the DTM within all the study area the existing DTMs were
mosaicked. Taking into account that the orthorectification of
areas with intense anaglyph is usually problematic, it was
considered necessary to use a highly accurate DTM in the hilly
region. Thus in this specific part of the image it was preferable
to use the DTM which covers this specific region and is
characterized by a better accuracy.
The process of mosaicking the DTMs consisted of three tasks.
Initially, the hilly part was extracted from the DTM that covers
all the study area. The extraction was made to the base of the
mountain along points that were found to have about the same
elevation in both DTMs. The second task involved changing the
grid-size of the DTM of the hilly part (20 m) to fit the grid-size
of the other DTM (25 m) through the process of resampling. It
was also necessary to cut the DTM of the hilly part near the
points mentioned above, so as to have a small overlapping with
the other DTM. Finally, the DTMs were mosaicked and as for
the overlapping area it was chosen to take the elevation values
from the DTM of the hilly part, with the better accuracy.
3.2.2 Refinement of the RFCs: The vendor-provided RF
coefficients can be improved to provide a more accurate
solution. Based on the known coordinates of the GCPs an affine
transformation can be applied using as initial values the vendor-
provided RFCs (Di et al., 2003).
187
3.2.3 Orthorectification in three parts of the image: The
orthorectification processing was applied three times for three
parts of the image. Due to the fact that images with an intense
anaglyph ~~ demand special attention during their
orthorectification, it was preferable to process the hilly part
separately. In order to have a good distribution of GCPs the
image was also processed separately as far as the west and the
east part were concerned.
In this way three orthorectified image parts were produced
which were finally joined together through mosaicking. The
same methodology was carried out for both the panchromatic
and the multispectral image and led to an improved accuracy of
about 3.5 m.
Image parts No. RMS
Control Check Control Check
West part 23 10 3.7m 3.1 m
East part 51 15 3.9m 3.7m
Hilly part 29 10 29m 29m
Table 2. The number of points and RMS errors of the
orthorectification of each part of the image
3.3 The results of the orthorectification
There are many factors that can influence the orthorectification
processing and reduce the accuracy of the orthoimages. In this
case the reasons can be found in the insufficient accuracy of the
initial DTM, of the vendor-provided RFCs or in the different
sources of GCPs (GPS and orthophotos). Using the
methodology mentioned helped facing these problems and
minimized the RMS error to a sufficient extend.
In Table 3 the results of the orthorectification are listed
according to the data being used and the great improvement of
the accuracy is shown.
DTM RFCs RMS
Initial DTM Before the 35m
refinement
After the
Image
Whole image
Whole image | Synthetic DTM s 10m
refinement
Three parts of Synthetic DTM After the 35m
refinement
the image
Table 3. RMS errors of the orthorectification according to the
data used
4. FUSION
4.1 The fusion of images
In 1999 Wald defined data fusion as a formal framework that
expresses the means and tools for the alliance of data
originating from different sources; it aims at obtaining
information of greater quality; the exact definition of “greater
quality” will depend upon the application. Image fusion aims at
the generation of a single image from multiple image data for
the extraction of information of higher quality (Pohl, 1999).
Within this framework this project focused on a type of image
fusion, the image sharpening. In image sharpening, higher
spatial resolution panchromatic data is fused with lower spatial
