ul 2004
lishing
lishing
in the
Wuhan
em for
Digital
ecord,
[FILM
IMAGE FUSION OF LANDSAT AND KOREAN SATELLITE KOMPSAT
S.H Han, J.M Kang"
“Dept. of Civil Engineering Cheonan National
Technical College | shehan(g)entc.ac.kr
b x ~ . . . = . . . .
Dept. of Civil Engineering, Chungnam National University, jmkang@enu.ac.kr
Commission I, WG 1/3
KEY WORDS : Image fusion, Spatial resolution, Wavelet fusion, HIS
ABSTRACT :
Korea is providing images with the resolution of 6.6m level as launching multi-purpose satellite, Arirang satellite, in Dec.
1999. This study analyzed advantages and disadvantages of image fusion methods as applying fusion techniques to panchromatic
images of high resolution of Arirang satellite and multi-spectral images of LANDSAT as test images. Analysis showed that the
wavelet image fusion technique is good for a variety of test fields.
1. BACKGROUND AND GOALS
This study aims to create components for fusion
techniques for spatial information of high resolution images
and spectral information of multi-spectral images. General
purpose of image fusion is to produce images that can be more
casily analyzed as enhancing spatial resolutions of low
resolution images with spectral data. This study carried out
creation of components as applying the method to fuse spectral
resolution of multi-spectral images from multiple wavelength
ranges with inferior spatial resolution and spatial resolution of
the latest panchromatic images of high resolution such as
IKONOS. Among fusion methods for fusing multi-spectral
images of low resolution and high resolution, the most common
methods are IHS(Intensity Hue Saturation), HPF(High Pass
Filter), PCA(Principal Component Analysis) and Wavelet
Fusion that has been recently researched a lot(Kim, 2000).
The goal of this study is to develop a system through
development of modules using Visual Basic and Visual C++
and to define algorithm for IHS and Wavelet fusion methods.
Moreover, this study analyzes compatibility of satellite fusion
methods as applying a developed system to Arirang
(KOMPSAT; KOrean Multi-Purpose SATellite) and identifying
problems.
2. RESEARCH
2.1 Preparation of Test Images
This study prepared star images(500*500) by R.G.B. band of
Landsat and black and white images(1923*2124) of
KOMPSAT(EOS) along the Hangang riverside. Land covers
varies on images so that test images were selected as Landsat
319*319 and KOMPSAT 1335*1335. Raw images were
normalized for easy processing. Figure 1 shows the images.
Window parts of images indicated test areas. To enhance
readability of images, histograms were analyzed and
normalized by band. Test areas were defined on the basis of
KOMPSAT images with high resolutions.
2.2 Geometric Correction of Images
For absolute comparison between IHS and Wavelet Fusion
method, both images were sized enough to satisfy all test
conditions. Landsat images were generated through fusion of
R,GB bend. Then, the geometric correction was applied to
Landsat images as referring to KOMPSAT as reference images.
Geometric correction of images used ERDAS Imagine 8.4.
To this end, the reference points were set to 6 points with good
luminance on images and the check points were set to 5 points.
Figure 1 Testing area. (KOMPSAT B&W)
The table below explains the process for geometric
correction. If overall errors on the reference point and the check
point were within 1 pixel, they were accepted as allowable
errors. In preparing test images in this study, the errors was
Table 1 Residual and RMSE of check points when Geometric
correction
Check X Y RMSE Contrib.
point residual residual (pixel)
7 -0.071 0.036 0.080 0.630
8 -0.001 0.195 0.195 1.538
9 0.031 -0.038 0.049 0.386
10 0.009 -0.134 0.134 1.059
11 0.004 0.175 0.175 1.385
12 -0.009 0.027 0.029 0.225
Avg.(ABS) 0.020 0.101 0.110 0.870
about 0.13 pixel for geometric correction. Table 1 describes the
results of correction processing. It was required to satisfy limit
requirements to apply each fusion method. In particular, for
wavelet method, image sizes had to be limited to 2" so that
images were sized to 512*512 by adjustment. This study
identified and applied the method to most effectively extract
images with the sized of 2".