ons
ns,
V to
ges
the
000
[HS
tral
ults
of a
”
OS
ale
rea
ing
6:
at
eas
ing
nal
ige
any
Figure 2 - High-Pass Filter used (Godoy Jünior, 1993).
2.2 IHS transformation
This technique, frequently used to integrate images obtained
from different sensors, allows to substitute the Intensity
channel of the multispectral component in IHS domain, by a
better spatial resolution band.
In this paper, the IHS transformation was applied to generate
the synthetic image resulting from the fusion of the
multispectral SPOT image with the filtered panchromatic
SPOT image. The following procedure was used, aiming at
preserving the spectral characteristics presented by the urban
areas in the original multispectral image, without losing the
spatial resolution of the panchromatic image previously
filtered:
* In RGB domain, the histograms averages of the three
multispectral SPOT image bands were read (Figure 3). This
multispectral component had been previously resampled in
the registration operation with the panchromatic SPOT
image.
* These averages were equalized to 128.
* The transformation of this equalized image to the IHS
domain was performed.
e The IHS image statistical parameters were read, so that the
component I average was 127.45.
e The statistical parameters of the filtered panchromatic SPOT
image were read and its average was 40.61.
This average was equalized to the multispectral SPOT image
band I average, which would in turn substitute it.
e The statistical parameters of the filtered panchromatic SPOT
image were read and the equalized average was equal to
126.58.
* In the IHS domain, the component I of the multispectral
SPOT image was replaced by the panchromatic SPOT image
whose average was equalized.
* This synthetic image was transformed from the IHS to the
RGB domain.
* The statistical parameters of the new RGB image were read.
They had the following values: Band R=125.00; Band
G=125.84; Band B=126.56.
735
e From each band average, the necessary values were
subtracted in order to return to the original resampled
multispectral SPOT image averages, before the equalization.
e The final synthetic image statistical parameters were read,
with the following average values for each band (Figure 4):
Band R=39.98; Band G=31.83; Band B=41 81.
3. RESULTS
The high-pass filter application, previously described, allowed
to refine the original panchromatic image visual quality. The
filtered image enhanced the river banks, as well as the intra-
urban features of Säo José dos Campos city and also the roads,
bridges and railways.
The comparison between the histograms of the resampled
multispectral SPOT image (Figure 3) and the synthetic image
(Figure 4) enables one to see that through the described
processing a final image was obtained with spectral
characteristics very similar to those of the original
multispectral component. However, the synthetic product
appeared with a slight alteration in its spectral characteristics,
thus originating some shape and size changes of certain
features, specially in areas of dense vegetation. This was due to
the substitution of the multispectral image channel I by the
filtered panchromatic band, as it has a low spectral response to
the close infra-red and therefore, in the synthetic image, areas
with vegetation as well as other targets which have a high
degree of reflectance in the infra-red band, appeared with a
somewhat different spectral response as opposed to that of the
original multispectral image bands. That results in a certain
degree of spectral information loss in those areas. On the other
hand, the intra-urban areas were considerably improved.
4. CONCLUSIONS
Based on the results obtained, the following conclusions can be
drawn:
A. The edge enhancement obtained through the high-pass
filtering of the original panchromatic image, favoured a
better discrimination as well as the digitalization of several
linear and punctual features which did not appear well
defined in the original image, proving to be very efficient
to obtain more enhanced synthetic bands.
B. The merging, in the synthetic image, of the spatial
resolution of the filtered panchromatic image with a
spectral resolution very similar to that of the original
multispectral component, allowed a better identification
and extraction of targets such as industrial areas, urban
areas limits, street sections and other intra-urban features.
C. It is interesting to emphasize that in this methodology the
original multispectral bands, without altering the contrast
nor any kind of enhancement, were used. The original
bands histograms, as well as those of the translated original
bands to the average value used for the equalization (128)
did not exceed the superior limit of 255 to avoid
information loss.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996