The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008
2. IMAGE FUSION TECHNIQUES
2.1 Smoothing Filter-based Intensity Modulation
SFIM is spatial domain fusion method based on smoothing low-
pass filters. It is defined as:
Jl *b h
B SFIM - B M
(1)
Where B L is LSR image, B H is HSR image, B M is a simulation
image of LSR image, and it derived from B H image using an
average low pass filter.
SFIM can be understood as a LSR image modulated directly by
high spatial frequency information and the fused image is
independent of the contrast and spectral properties of the HSR
image. It is critical for the selection of a low pass filter kernel to
generate B M image. The minimal filter kernel size is decided
based on the resolution ration between the higher and lower
resolution images (Liu,2000). For example, the minimal filter
kernel size should adopt 4x4 filter window for fusing LSR
image with HSR image of Quickbird or IKONOS. SFIM-fused
images have fine spectral information preservation with the
LRS images, but the edges between different features are
blurred in SFIM-fused productions (Liu,2000).
2.2 Wavelet Transform
More recently, Discrete WT (DWT) has started playing a role
in image fusion. In general, DWT consists of wavelet
decomposition (Fig. 1) and reconstruction (Fig. 2). There are
following steps in image fusion based on DWT.
1) Selecting biorthogonal wavelet bases used in wavelet
decomposition and reconstruction. 2
W r , W r " , W; and W r D (Fig. 1) are approximation
coefficients, horizontal coefficients, vertical coefficients and
diagonal coefficients, respectively.
3) The low frequency sub-image of LSR image and the high
frequency sub-images of HSR image are selected to generate
the fused image by wavelet reconstruction (Fig. 2).
Low- pass
К
к
к
-►
к
Low- pass
f+ x » sub- image ►
Ж
h
High- pass
Low- pass
V
+/‘
A
Fused
image
w:
<p
►
K
W D
<p
K
S sub- image ►
If
JL
High- pass
High- pass
Figure 2. The wavelet reconstruction of two-dimensional DWT
The key issues in DWT are selections wavelet bases,
decomposition levels and replacement sub-images. Different
selections form different fusion models. Depending on the
optimization control, fusion methods based on DWT can better
preserve image spectral properties than IHS transform and BT.
However, because of more complicated and time-consuming
processing and critical requirement for the co-registration
accuracy, these techniques are less popular than IHS transform
and BT for remote sensing applications which prefer fast
interactive processing and real time visualization (Liu,2000).
2) The selected wavelet bases are applied to decompose LSR
and HSR images. After decomposition at any level, the low
frequency sub-image (commonly termed “approximation”
coefficients) is passed to the next decomposition. High
frequency sub-images (termed “horizontal”, “vertical”, and
“diagonal”) are retained for reconstruction.
2.3 Brovey Transform
One of widely used image fusion methods is BT based on
chromaticity transform and RGB space transform. It is a simple
and efficient technique for fusing remotely sensed images. BT
is defined as:
image
Low- pass
Low- pass
h
sub-image
Г к
w,
H
High- pass
Low-pass
К
wf
► hy > sub-image
High- pass
High- pass
Figure 1. The wavelet decomposition of two-dimensional DWT
R BT ~ j X Bh » G B t ~ j x B H
n r R + G + B
D BT ~ J * D H'> 1 ^
(2)
Where R, G and В are three bands LSR images.
Only three bands LSR images are involved in standard BT.
Thus, there are different choices when the total number of LSR
images is more than three. For example, there are four different
choices in the case of fusing four bands LSR images of
Quickbird or IKONOS. To solve the problem, Modification BT
(МВТ) fusion method based on the principle of BT is
introduced in this paper. МВТ is defined as:
1228