1253
AN IMPROVED IHS IMAGE FUSION METHOD
WITH HIGH SPECTRAL FIDELITY
Wen Dou a , Yunhao Chen b
a Department of Geographic Information Engineering, Southeast University, Nanjing, Jiangsu, 210096 P. R. China
- douw@seu.edu.cn
b College of Resource Science, Beijing Normal University, Beijing, 100875 P. R. China - cyh@ires.cn
Commission WG VII/6
KEY WORDS: remote sensing, image fusion, histogram match, IHS
ABSTRACT:
Image fusion is a critical issue for remote sensing, and many algorithms have been developed. Image quality assessment of fused
image might provide comparison between fusion methods, but the conclusion is not so general because different test images would
lead to different assessment results. The paper studies on the relationships between image fusion methods aiming to reveal the nature
of various methods. By doing so, we could compare the performance of spatial enhancement and spectral fidelity from mathematical
form of image fusion processes.
1. INTRODUCTION
Methods based on the Intensity Hue Saturation (IHS) transform
are probably the most popular approaches used for enhancing
the spatial resolution of multispectral (MS) images with
panchromatic (PAN) images (Tu, T.M., et al.,2004). The IHS
method is capable of quickly merging the massive volumes of
data by requiring only resampled MS data. Particularly for
those users, not familiar with spatial filtering, IHS can
profitably offer a satisfactory fused product.
The main concept of the IHS method is based on the
representation of low-resolution MS images in the IHS system
and then substituting the Intensity component I with the PAN
image. However, IHS and other so-called “component
substitution” methods would introduce spectral distortion into
the resulting MS images, appearing as a change in colors
between compositions of resampled and fused multispectral
bands. Such methods take redundant information of the PAN
and MS imagery as the basis of image fusion, and hypothesize
that PAN image and the Intensity component of the MS image,
which is retrieved based on the RGB color model, contain
almost the same information. That means PAN is taken as the
high resolution intensity component of the high resolution
multispectral data. Based on such hypothesis, spatial detail is
the difference of PAN and the low resolution I component, and
is injected into the MS image by substituting the I component
with the PAN image.
Unfortunately, it is impossible to construct I component
containing same information as PAN image. That means spatial
detail could be far from the truth when I component is
constructed in an improper way. Therefore, introduction of
spectral distortion is partly due to the construction of spatial
detail.
histogram matching should be implemented to make the PAN
image has the same average and standard deviation with the low
resolution I component as (1).
= —-—I- fig (i)
where fh is high resolution intensity component, and
fJLj are average of PAN and I respectively, and tT/rjf and
O’i are standard deviation of PAN and / respectively.
It seems that the process is reasonable to make 1 and PAN
comparable. However, an image vector space V is introduced
to analyze histogram matching process, in which a single band
image could be represented as a vector. For any vector £ and
TJ in V, dot product is defined as
<■ >= (2)
where is covariation of ^and TJ ■
It is easy to prove that the vector space y with this operation is
an inner product space. Module of vector and angle of two
vectors o' is
m= P)
2. ANALYSIS ON HISTOGRAM MATCHING PROCESS
For the IHS method, I component is constructed by the average
of R,G,B band. Before / is substituted with PAN image,