A NEW DATA FUSION METHOD FOR IMPROVING CBERS-1 IRMSS IMAGES BASED ON 
CCD IMAGE 
Z.R. Qi 
Centre for Aero Geophysics and Remote Sensing, Ministry of Land and Resources, 29, College Road, Beijing, 
China, gizr@cugb.edu.cn 
KEY WORDS: CCD, Fusion, High Resolution, Registration, Pixel, Spectral 
ABSTRACT: 
A new data fusion method is proposed in this paper to improve 
IRMSS image with lower spatial resolution is correspondence 
resolution after precise co-registration. For each pixel in the 
brightness values in the CCD image is computed. The difference 
CBERS-1 IRMSS images based on CCD image. Each pixel in an 
with a group of sub-pixels in a CCD image with higher spatial 
[RMSS image, the average value of its co-registered sub-pixel 
between the pixel brightness value in the IRMSS image and its sub- 
pixels average value was added to every sub-pixel value in the CCD image. This procedure is applied to each pixel to produce a data 
fusion image. The fusion image had more spatial details than the IRMSS image and had similar spectral information to the IRMSS 
image. The test indicated that the new data fusion method was a superior fusion technique not only enhancing the spatial details of 
IRMSS image but also keeping the fidelity to the image spectral properties. Different from the other data fusion methods, which 
need more than one band low resolution image to be merged with one high resolution image, the new method can perform data 
fusion for an individual low resolution image and it can be used 
other higher resolution band image. It is a new, very simple and 
for enhancement of the thermal infrared band 6 of Landsat with the 
practical technique. It is potentially useful for enhancing the spatial 
details and preserving the spectral fidelity of the low spatial resolution images. 
1. INTRODUCTION 
China-Brazilian earth resources satellite 1(CBERS-1) is 
equipped with three types of sensorsCisuch as CCD camera 
Infrared Multi-spectral Scanner(IRMSS) and Wide Field 
Imager(WFI), and provides three types of remote sensing image 
data (see table 1). Because of lower spatial resolution, CBERS- 
| IRMSS images have not been widely used in application. 
Data fusion technique can be used for improving the spatial 
resolution and enhancing the spatial details for IRMSS images 
with CCD image, which is potential useful and significant for 
developing IRMSS image application market. 
There are many methods for data fusion at the present, and they 
can be divided into two methods: spectral component 
substitution techniques and spatial domain techniques. The 
former replaces a spectral component of the multi-spectral 
images by the radio-metrically-adjusted pan-image such as hue- 
intensity-saturation (HIS) colour spatial transformation and 
principal component analysis (PCA) (P. S. Chavez et al, 1991: 
C. Conese et al, 1992). The latter adds high resolution 
information from the high resolution image to all the low 
resolution spectral bands by using various deterministic or 
statistical predictors, such as high pass filter (HPF) (C.Conese 
et al, 1992; R. A. Schowengerdt, 1998; V.T. Tom et al, 1985), 
wavelet transform (WT) (H. Li et al, 1995; B. Garguet-Duport 
et al, 1996). Some researches combine two techniques, such as 
combination of multi-resolution wavelet decomposition and 
HIS transform for the fusion of high resolution pan-image and 
multi-spectral images (Jorge Nunez et al, 1999). Price uses 
panchromatic gray value to estimate single multi-spectral image 
values as a result of the highly-correlated statistical analysis of 
SPOT panchromatic channel (0.51-0.73pm) and TM 1(0.45- 
0.52pm) ITM 2(0.52-0.60pm), so this algorithm is confined to 
the images of highly-correlated statistical analysis (John C. 
Price, 1987 and 1999). Zhukov etc. utilize the class information 
of high resolution images and PSF (point spreading function) to 
un-mix the low resolution images, but this algorithm needs 
multiple high-resolution images so as to classify them (Boris 
Zhukov et al, 1999). Ryuei Nishii etc. employ a multivariate 
normal distribution for the seven TM band values to predict the 
value of band 6 by the conditional expectations (Ryuei Nishii et 
al, 1996). Jiao etc. present a data fusion algorithm based on 
scale-transformation at spatial domain (Jiao et al, 2002). 
Table 1. Some technique parameters of CBERS-1 sensors 
  
  
  
  
  
  
Sensor Bands/ Lim Resolution Image width 
/m in equator/km 
B1: 0.45-0.52 20 113 
B2: 0.52-0.59 20 113 
CCo B3: 0.63-0.69 20 113 
camera |.p4:0,77-089. | 20 113 
B5: 0.51-0.73 20 113 
B6: 0.50-0.90 78 120 
B7: 1.55-1.75 78 120 
IRMSS B8: 2.09-2.35 78 120 
B9: 10.4-12.5 156 120 
B10: 0.63-0.69 | 258 890 
WEE | B11:0.77-0.89 [ass 890 
  
  
  
  
All these methods described as above can improve the spatial 
resolution for lower resolution images, but they cannol 
completely restore the lower resolution images, i.e., they will 
not preserve the spectral fidelity of lower resolution images 0 
some extent after data fusion processing. For image processing, 
image spectral analysis and interpretation, preserving spectral 
fidelity is more important than enhancing the spatial details and 
improving the spatial resolution for the lower resolution image 
In order to integrate and process CCD and IRMSS images, and 
to mine the multi-spectral properties in CBERS remote sensing 
data, it is necessary to develop a data fusion method not only 
910 
Int 
for 
fid 
Ge 
col 
sul 
reg 
res 
val 
Th 
pb 
no 
fid 
im 
pb 
bri 
pb 
eq 
in 
aft 
im 
or 
Fi 
Be 
im 
in 
foi 
its 
co 
bri 
ha 
im 
ev 
pr 
wi 
im 
sal 
Wi 
Th 
(1) 
val 
res 
on 
sul 
fur 
Ol 5