International Archives of Photogrammetry and Remote Sensing, Vol. 32, Part 7-4-3 W6, Valladolid, Spain, 3-4 June, 1999
112
Image
Original
AIF
IHS
bands
image
LISS-2
P
89.6
89.6
91.1
green
G
21.7
21.0
26.0
LISS-3
P
32.1
32.0
33.0
red
G
13.0
12.6
14.8
LISS-4
P
100.6
100.8
99.4
NIR
G
26.1
23.6
20.1
Table 1. Statistics of differences between original and
degraded image bands (IRS-1C).
Comparison of the correlation coefficients leads to similar
results (see Table 2). It can be clearly seen that the AIF image
bands correlate highly with the original data, whereas the
correlation between the IHS merged image and the original
image is significantly lower. As expected, the difference in the
near infrared band is highest as it is least correlated with the
panchromatic image.
Bands
AIF
IHS
LISS-2
0.99
0.92
LISS-3
0.99
0.97
LISS-4
0.97
0.85
Table 2. Correlation coefficients between AIF and IHS
merged images with the original ones (1RS-1C).
3.2. Urban case study
In order to evaluate the AIF on very high resolution data, a
second study was performed on airborne scanner data. The
image was acquired by a Daedalus AADS 1268 ATM (Airborne
Thematic Mapper) scanner in August 1991 and covers an area in
the south of Vienna, Austria. The spectral resolution of this
scanner comprises 11 bands. For the case study only the 6 bands
that correspond to the spectral bands of Landsat TM were used.
The scanner data were geocoded to the Austrian reference
system (Gauß-Krüger M34) with a spatial resolution of 6m.
The focus of this study was an analysis of AIF results compared
to multispectral images observed with the high resolution of the
panchromatic image. In order to perform this comparison the
ATM image (denoted by org in Table 4) was degraded to a
resolution of 18m (denoted by deg in Table 4) and an artificial
panchromatic image was computed by averaging bands 1 to 4 on
the 6m pixel basis. Next, an AIF was performed on the degraded
image bands and the panchromatic image resulting in an
artificial image stack with a resolution of 6m. This image stack
was then compared to the original image data.
As it can be seen in Table 3, the global mean values of all three
images (original, degraded and fused) are stable, thus no spectral
distortion occurred during the fusion process. Again the standard
deviation of the degraded and the fused image is lower than the
one of the original image. For more detailed analysis the
correlation between the three image stacks were computed (see
Table 4). Whereas the correlation between the original image
and the degraded image lies between 0.82 and 0.85, the
correlation between the original and the AIF is significantly
higher except for band 5. This effect might be due to the fact
that the correlation between the multispectral and the pan
chromatic image is lowest for band 5.
Image
Original
Degraded
AIF
bands
image
image
image
ATM 1
P
85.3
85.4
85.1
blue
G
27.3
19.5
19.9
ATM 2
P
56.3
56.3
56.2
green
G
19.5
14.1
14.5
ATM 3
P
51.2
51.2
51.0
red
G
24.2
18.0
18.5
ATM 4
P
67.5
67.6
67.7
NIR
G
21.7
15.7
16.0
ATM 5
P
69.2
69.1
69.3
NIR
G
27.3
21.0
20.0
ATM 6
P
71.6
71.5
71.4
MIR
G
31.7
23.7
23.6
Table 3. Statistics of differences between original and
degraded image bands (ATM).
Bands
org - deg
org - AIF
deg - AIF
ATM 1
0.82
0.94
0.87
ATM 2
0.83
0.94
0.88
ATM 3
0.84
0.94
0.89
ATM 4
0.82
0.93
0.82
ATM 5
0.85
0.84
0.87
ATM 6
0.84
0.89
0.88
Table 4. Correlation coefficients (ATM).
A visual comparison shows the benefits and limitations of the
AIF (Fig. 4). Compared to the degraded image, the fused image
is clearly sharper and shows no blockiness. Although most of
the object structures seen in the original image will be found in
the fused image as well, some of the smaller objects are blurred
(e.g. garden area in the upper left part of the image, single
buildings in the village in the lower part). In these cases the
single objects are too small to be reconstructed from the
degraded image.
3.3. Forest case study
In a third case study the use of AIF was tested for a forest
application. The objective was the derivation of a forest mask
from Landsat TM imagery in combination with a panchromatic
SPOT image in an alpine region. The two images were acquired
on August 18, 1992 and August 17, 1992 respectively. The area
is located in the center of Austria, in the Dachstein region,
Styria (centered on 13°40’E / 47°20’N). Both images were