1129
AN AUTO-ADAPTIVE INFORMATION PRESERVATION FUSION
METHOD FOR SAR AND MULTISPECRAL IMAGES
H. Sun 3 , B. Pan b , Y. Chen 3 ’*, J. Li 3 , L. Deng 3
d College of Resources Science & Technology, Beijing Normal University
b School of Remote Sensing and Information Engineering, Wuhan University
KEY WORDS: Fusion; SAR; Multispectral; Wavelet; Auto-adaptive
ABSTRACT:
Due to the special imaging model of SAR, its image includes certain special information that unavailable in multispectral image,
however, the interpretation of SAR image becomes more difficult than visible image. Therefore, the fusion of SAR image and
multispectral image is primarily for the purpose of making better use of SAR image information. For this application, the auto-
adaptive information preservation fusion method based on wavelet transform was proposed. It automatically adjusts weight
coefficient according to the information capacity of SAR image and multispectral image, and then auto-adaptively adjusts fusion
result, avoiding a single and ponderous fusion strategy for the whole image. Experiment results show that the novel method can
better preserve SAR image texture and structure information, especially the unique information in SAR image, meanwhile, the novel
method as well as better preserved spectral information with good results.
1. INTRODUCTION
The remote sensing images acquired by various single sensors
have some limitation and differences in the respect of geometry,
spectrum and spatial resolution. Therefore, it is necessary to
fuse or synthesize the information contained in the images of
multiple sensors, in order to enhance the distinguishing
reliability and interpreting ability of the images (Li,2003), and
prepare for the classification, target recognition and other more
latter applications.
The traditional processing method is that fusing multispectral
image of low spatial resolution and panchromatic image of high
spatial resolution and gaining result image which not only
owing higher spatial resolution in panchromatic image but also
obtaining spectral information in multispectral image. As
respect to application, the processing method might enhance the
interpreting ability of original images. Nevertheless, different
applications require the fused images not just eclectically
contain the detail information in panchromatic image and
spectral information in multispectral image. As for the
application of feature extraction, mapping, the detail
information of image is normally more important; while as for
the application of classification, it is normally more important
to focusing on the accuracy and integration of the spectral
information (Deng,2005).
For the fusion of synthesize aperture radar (SAR) image and
multispectral image, SAR image is regarded as main body to
contain the detail information better, and the multispectral
image is used to supply additional spectral information. As the
imaging mode of SAR is active in microwave band, SAR can
penetrate clouds, rain and snow, and bear the ability of round-
the-clock working and in some sense penetrating earth’s surface
(Woodhouse, 2004). Therefore, SAR sensor might acquire
information that unavailable for visible or infrared sensor.
However, as the special imaging way SAR has, the
interpretation of SAR images is more difficult than that of the
visible ones, it is necessary to fuse SAR image and
multispectral image, in order to better utilize the special
information SAR image contained.
The traditional fusion methods including Intensity-Hue-
Saturation (IHS) transform, Principle Component Analysis
(PCA), Wavelet Transform and High Pass Filter (HPF), etc.
And more and more advanced methods based on the traditional
fusion methods have proposed (Li, 1994). Most of these
methods focus on preserving spectral information or eclectically
preserving information between spatial detail one and spectral
one, which resulting the fused image usually only advance the
visible effect. But scarce methods regard panchromatic image
as main body, aiming to maintaining as much as detail
information.
In this paper, from the aspect of application-oriented, we
proposed a novel fusion method to synthesize SAR image and
multispectral image. This novel method is based on wavelet
transform, which we called auto-adaptive information
preservation fusion method. It not only better preserves spectral
information in multispectral image, but also, more importantly,
preserves special detail information in SAR image.
2. THE TRADITIONAL FUSION METHODS BASED ON
WAVELET TRANSFORM
An image can gain (37+1) components by 7-level two-
dimension discrete wavelet transform, one of which is low-
frequent approximate component, noted as Id , owing major
energy of original image, the other 37 ones, noted as
H[ , H' 2 and H[ respectively, are high-frequent detail
Corresponding author.
Email address: cyh@ires.cn