2004
the
spot
{ric
sen,
cial
nate
the
dis
nost
etric
and
| the
is of
the
IS
that
over
ntrol
jents
area.
and
ntrol
not
ss of
yd of
d on
able
n arc
osest
es of
value
es of
value
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B-YF. Istanbul 2004
In this project, Bicubic interpolation is selected, because this
method sharpens the image and smoothes out noise and
simultaneously the loss of image information is eliminated.
3.5 Results
The expected results of the process of rectification should be
near to sub pixel and this is achieved in this project, as it is
shown to the below tables:
X residual Y residual RMS Error
(pixel) (pixel) (pixel)
Control Point 0,6166 0,6503 0,8962
Check Point 0,5944 0,6139 0,8545
Table 2. The rectification results of the panchromatic image
X residual Y residual RMS Error
(pixel) (pixel) (pixel)
Control Point 0,7526 0,4541 0,8450
Check Point 0,7835 0,4715 0,9145
Table 3. The rectification results of the multispectral image
3.6 Check of the produced orthoimages
Finally, the resulted two orthoimages are tested in order to
assess their accuracy, by measuring and comparing the image
coordinates of control points to the known coordinates of the
available orthophotos. The chosen 20 points are well distributed
all over the area. The standard deviation is 8,9m for the
panchromatic image and 4,7m for the multispectral.
4. FUSION
It is often desirable to simultaneously require high spatial and
spectral analysis in a single image. This is accomplished with
the process of fusion. Fusion combines data from different
sensors with dissimilar resolution and provides images with
increased interpretation capabilities.
The images have to be rectified in the same reference system,
cover the exactly same area and have the same dimensions (the
same number of pixels / row and pixels / column), in order to be
fused. Firstly, it is necessary to registrate the low-resolution
image on the high-resolution image, so as to be possible to
compare these two images pixel by pixel. Moreover, it is
essential that the images have been orthorectified for the more
accurate pixel-by-pixel corresponding, especially in the case of
mountainous areas.
41 Preparation of the images
First of all the low-resolution orthorectified image is registered
to the high-resolution orthorectified image, for the absolute
coincidence of pixels. It is also necessary to define the exact
area of study in the two images; with view to obtain the corner
pixels of the two images the exact same cartographic
coordinates.
After these preparations the two images are ready to be fused
with the principal components analysis (PCA).
141
4.2 The principal components analysis (PCA)
In fusion with the PCA technique, the new band PCI can be
replaced by the panchromatic. This is possible because they are
considered to have the same spectral characteristics. By
inversion of the new principal components the result is one
synthetic image, which maintains the spatial characteristics of
the panchromatic image, and at the same time has the spectral
information of the multispectral. That is a synthetic image with
spatial resolution 10m.
Figure 4. The synthetic image
4.3 Evaluation of the spectral and spatial quality of the
synthetic image
In order to be more reliable and useful the synthetic image it is
important to valuate its spectral and spatial quality. The
comparison of the synthetic image to original multispectral is
accomplished under some certain conditions. The two
compared images must have the same spatial analysis and their
spectral information must be identical. That's why the synthetic
image should be degraded to the analysis of the original
multispectral, and the all band histograms of the synthetic
image must be matched with the histograms of the original
multispectral image.
For the spectral quality the following criteria are used:
1. Standard deviation, mean, correlation coefficient
(Wald et.al., 1997).
The NDVI index (Tsakiri, 2001)
D
