ite 
ad 
art 
nd 
t" 
ce 
al, 
on 
es 
ser 
nd 
C - 
ne 
hn 
rra 
on’ 
the 
Wa, 
ine 
Hs. 
ime 
ncy 
the 
iN) 
AN EFFICIENT METHOD FOR SATELLITE IMAGE MATCHING AND MANAGEMENT 
Jeong-Ho Park®, Jae-Ho Choi’, Kyung-Ho Choi" 
“Telematics Contents Research Team, Telematics Research Division, ETRI, DaeJeon, KOREA 
jhpark@etri.re.kr 
Division of Electronics and Information Engineering, Chonbuk Nat’l Univ, Chonbuk, KOREA 
wave(@moak.chonbuk.ac.kr 
Commission VI, WG VI/4 
KEY WORDS: Image matching, Wavelet transformation, Satellite Image, GCP chip, Energy Analysis 
ABSTRACT: 
A method for matching two different images is discussed in this paper. Image matching is to match an image with other image that 
may has different properties. In the matching, generally the images are the same in appearance, may contain quite different contents. 
If there is no noise, identical pixel value and unchanged edges in those, image matching process can be operated by very simple 
model. However, natural images to be matched in the most application do not have such an ideal circumstance. This paper proposes 
a compression method for archiving satellite images and matching scheme which is to match GCP images with a raw satellite image. 
The proposed method is based on wavelet transform, not required any pre-processing such as histogram equalization, noise reduction. 
1. INTRODUCTION 
Satellite images are material of great interest for many 
applications such as minerals management, investigation of 
environmental change and meteorology, etc. However, 
management of satellite images requires massive resources in 
terms of storage and data transmission. Therefore, viewed in the 
storage management and the fast transmission of data through 
network, techniques that are able to reduce the amount of 
satellite images are highly desirable(Ghassemian, H., 2001, 
Mittal, M.L., 1999). 
On the other hand, in order to use the satellite images in many 
applications, geometrical correction of the images is required 
essentially. One of methods for the correction is to utilize GCP 
chip images. GCP can be acquired from the chips, and then the 
correction will be effectively processed if DB for the chips was 
established. The first step of correction using GCP is to get 
GCP images by matching the chips to a raw image which will 
be corrected. Sometimes, the matching process requires a lot of 
time as well as it may cause matching error if spatial properties 
of the chip images have different ones of raw image. 
If satellite images are managed in storage before applying 
geometric correction to them, massive secondary storages are 
necessary. This paper proposes an efficient method for 
matching and management of the chip images and the raw 
images. All the raw images will be managed in compressed 
stream form, but can be expressed by a little parameter. 
Because variance of frequency bands can be one of useful 
parameters to analysis the spatial image, only the parameters 
and GCP will be stored into database before the matching 
method is applied. 
In this paper, an area matching method will be used for the 
matching, and the method is only to compare the parameters. 
Therefore, the proposed method is able to implement fast 
matching and is not required pre-processing to make similar 
circumstance as well. The main feature of our system is that all 
information of the raw images and the chips is managed in 
compressed stream by wavelet transformation, and chip 
matching in frequency band is accomplished. 
2. SPATIAL IMAGE AND WAVELET 
Transformation in image processing decreases correlation 
between pixels, and makes simple environment for the analysis 
of signal. Especially, because wavelet transform has not only 
these properties, but makes possible to decompose an image 
into multi-scale, wavelet is used in many application fields 
related to image processing. 
To analysis the relationship a spatial image and wavelet 
transform domain, we prepare an extreme image with definite 
direction as shown to figure 1. 
  
  
  
  
  
  
Figure 1. A spatial image with horizontal direction and its 
wavelet transform with 3 layers 
Figure 1 shows an extreme case, and never appears in the real 
image. However, from the figure, we can understand the fact 
that direction by variation of pixel value is exactly reflected in 
the wavelet transform domain. The other fact is that all 
significant coefficients only appear to each band according to 
the same direction. This means the change of brightness of 
pixels have great influence on energy distribution of the 
frequency bands. 
Of course, the real image consists of random change of pixel 
value, and it is not simple problem to reveal the relation 
- between variation of spatial image and energy of transform 
341 
domain. In this paper, simple model will be used to implement 
the proposed system instead of using complex mathematical 
structure.