The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part 87. Beijing 2008
I = Y^ W i MS i (5)
where MS', is the corresponding MS bands which covered with
PAN band, and w, is the weights of band i. For IKONOS, MS,
represent blue, green, red and near infrared band. For Landsat7
(ETM+), MS, represent green, red and near infrared band. And
for the other satellites, it can be referred to Table 1.
(2) Modulation of the spatial detail
Similar to GIF method proposed by Wang et al (2005), we
introduce a modulation coefficient (a) to modulate the spatial
detail. Then the method can extend traditional three-order
transformations to an arbitrary order and all the MS bands could
be fused at the same time. In this way, the equation (4) would
be rewritten as:
F(MS l )=MS k +a k -S=MS k +a t -(I mw -I) (6)
where MS k represent the MS bands which will be fused, / is the
intensity component that constructed according to (5), and
a k =MS k /I in order to keep the added spatial detail
proportional to their original values.
Take IKONOS images as an example, the difference among the
proposed method, the traditional IHS method, the fast IHS
method (FIHS) and the fast IHS method with spectral
adjustment (FIHS-SA) are shown in Table 2. In the proposed
method, the sensor spectral response has been considered
adequately and the spatial detail is injected into each band
discriminatively. There are several ways to obtain the weight
coefficients (Gonzâlez-Audicana et al,2006; Dou et al,2007).
However, they only consider the nominal spectral responses
which would be influenced by the on-orbit working conditions,
atmospheric effects or postprocessing effects. In this paper, the
PAN image is degraded to the same resolution as MS images by
means of low-pass filtering and subsampling. Assumed that
degraded PAN and MS bands satisfy the equation (3), a linear
regression algorithm is performed in order to estimate the
weight coefficients. Considering that there is a constant item
{other) in equation (3), an adjustment of mean value is required
to keep the global spectral balance.
Fusion method
Wj
W>2
VV?
w 4
a k
IHS
1/3
1/3
1/3
0
1
FIHS
1/4
1/4
1/4
1/4
1
FIHS-SA
1/12
1/4
1/3
1/3
1
Proposed method
w B
w G
w R
W NIR
MS/JI
Table2. Comparisons of different fusion methods
4. EXPERIMENTAL RESULTS
To evaluate the performance and efficiency of the proposed
method, experiments are carried out on IKONOS, Landsat 7
ETM+ and EO-1 ALI images respectively. For the experiment
on the fusion of IKONOS images, the original PAN and MS
images are first atmospherically corrected and then spatially
degraded to a resolution of 4 and 16 meter, respectively. The
performance of the proposed method is compared together with
traditional IHS method and three typical modified IHS methods
(IIHS method proposed by Xiao 2003; FIHS method proposed
by Tu,2004; IHS-WT method provided in ERDAS) both
visually and quantitatively. Part of image is extracted to
compare the visual effect of the fused images with reference
image (the original MS image). From Figure.2, it can be easily
seen that the fused image generated from traditional IHS
method has obvious colour distortion. The spectral quality of
FIHS fusion result has improved to some extent, but the colour
of vegetation area in top right comer is still changed. The fused
images generated from IIHS and IHS-WT methods keep good
spectral quality, but the spatial quality of them are not as good
as the other ones. The proposed method preserves almost all the
spatial details and minimizes spectral distortion. The fused
image generated from it is most similar to the reference image.
To quantitatively assess the spectral and spatial quality of the
fused images, some indices including bias, correlation
coefficient (CC), spatial correlation coefficient (sCC,), and the
universal image quality index (UIQI) are used. The bias refers
to the difference between the means of the fused and reference
images. The smaller the difference, the better the spectral
quality is. The CC between the fused and reference image
shows similarity between them. The sCC is proposed by Zhou
et al (1998). In the procedure, the PAN and fused images are
filtered with a Laplacian filter and the correlation coefficient
between the filtered images is defined as sCC. The high
correlation coefficients indicate that most of the spatial details
are injected during the merging process. The UIQI indicates the
spectral quality of the fused image (Wang et al,2002). The
bigger the value of UIQI, the better the spectral quality is. From
Table 2, we can find that except a litter smaller in CC index
than IHS-WT method, the proposed method has superior
performance than other methods in both the bias and UIQI
index, which means the smallest spectral distortion.
Furthermore, it is clear that the fused image from the proposed
method has a similar sCC in comparison to those generated
from IHS and FIHS method, which is much higher than those
from IIHS and IHS-WT methods. To sum up, the proposed
method has the best comprehensive performance.
In addition, the experiment results of Landsat 7 ETM+ and EO-
1 ALI images are shown in Figure 3 and Figure 4 respectively.
For better comparison, a subset of images and the fused results
are selected and displayed by using the same linear stretch
method. From the picture, it is noticeable that the fused images
from the traditional IHS method have obvious spectral
distortion, such as the airport runway and water in Figure 3 and
the vegetation area in Figure 4. Unlike the traditional IHS
method, the improved IHS fusion method proposed in this paper
generates fused images with both high spectral fidelity and high
spatial resolution. Moreover, all the MS bands (IKONOS (1-4),
ETM+ (1-5, 7)) have been fused at the same time by using the
proposed method.
1171