1159
THE STUDY ON IMAGE FUSION FOR HIGH SPATIAL RESOLUTION REMOTE
SENSING IMAGES
S.S.Han 3 ’ b ’ *, H.T.Li 3 , H.Y. Gu a ’ b
1 Institute of Photogrammetry and Remote Sensing,Chinese Academy of Surveying and Mapping, Beijing
100039,China - joinshanshan@163.com,lhtao@casm.ac.cn,haiyanl982709@126.com
b Institute of Surveying and Geography Science, Liaoning Technical University, Fuxin 123000,China
- joinshanshan@ 163 .com,haiyan 1982709@ 126.com,
Commission VII, WG VII/6
KEY WORDS: Remote Sensing, Image Fusion, Image Evaluation, Multiplication, Modified Brovey, High-Pass Filter, Smoothing
Filter-based Intensity Modulation
ABSTRACT:
Remote sensing images fusion can not only improve the spatial resolution for the original multispectral image, but also preserve the
spectral information to a certain degree. In order to find out the fusion algorithm which is suited for QuickBird images fusion, four
simple fusion algorithms, Multiplication (MLT), Modified Brovey (MB), High-Pass Filter (HPF) and the Smoothing Filter-based
Intensity Modulation (SFIM) algorithm have been employed for result evaluation. The study is based on a QuickBird sub-scene
covering different land use. Numerical statistical methods such as Bias of Mean, Correlation Coefficient, Entropy, Standard
Deviation and Average Grads are used to quantitatively assess the fused images produced using the above algorithms. The analysis
indicates that the SFIM-fused image has the best definition as well as spectral fidelity, and is the best in high textural information
absorption. Therefore it is suited for QuickBird image fusion best.
1. INTRODUCTION
Nowadays, remote sensing is developing to high-spectral
resolution, high-spatial resolution, and high-time resolution. But
as far as one and the same data is concerned, it is difficult to
obtain the image of high-spatial resolution and high-spectral
resolution at the same time. The information content of a single
image is limited by the spatial and spectral resolution of the
imaging system. Since the advent of the high spatial resolution
satellite images, the merging of multiresolution images has been
an important field of research. The fusion of remote sensing
images can integrate the spectral information of single sensor or
the information from different kinds of sensors (Couloigner,1998).
In order to improve the dependability for extracting remote
sensing information, and enhance the efficiency of using data.
Literature has shown a large collection of fusion methods
developed over the last two decades, such as the Multiplication
(MLT) algorithm, Modified Brovey (MB) algorithm, High-Pass
Filter (HPF) algorithm, the Smoothing Filter-based Intensity
Modulation (SFIM) algorithm (Liu,2000a), the Principal
Components Analysis (PCA), the Intensity-Hue-Saturation (IHS)
and so on. All of the above-mentioned methods can realize the
fusion of multi-spectral and high-resolution images, besides it can
improve the spatial resolution and preserve the spectral
information to a certain degree. For the moment, every method
has its own advantages and disadvantages. Even if we use the
same fusion to deal with the different images, we will get the
different effect. In this paper, we evaluate the effect and
applicability of different methods for high-resolution images
through comparing the staple fusion methods. It can offer the
reference to the fusion of high-resolution images.
2. FUSION ALGORITHMS
Data fusion provides several advantages (Ranchin,2003):
preservation of computer storage space; enhancement of
aesthetic and cosmetic qualities; improvement of spatial
resolution; and analytical improvements. Each reason for data
fusion relies on the following premise—for a data fusion
model to be effective, the merged images should retain the
high spatial resolution information from the panchromatic
(Pan) data set while maintaining the basic spectral record of
the original multi-spectral (MS) data (Carper et al,
1990).Many methods have been developed in the last few
years producing good quality merged images. This study
analyzes four current data fusion techniques to assess their
performance. The four data fusion models used include MLT,
MB, HPF, SFIM algorithms. The reasons of selecting the
above methods is mainly as follows: 1) They are all
mathematically similar, for example, they are all statistical-
based methods rather than color-related techniques; 2) They
are simple and easy to be used; 3) They can be performed
with any number of selected input bands, while some others
like HIS only allow a limited number of input bands to be
fused.
2.1 MLT Algorithm
The Multiplication model combines two data sets by
multiplying each pixel in each band of the MS data by the
corresponding pixel of the Pan data (Pohl.C,1997). To
compensate for the increased Brightness Values (BV), the
square root of the mixed data set is taken. The square root of
* E-mail: joinshanshan@163.com; Phone: 86 10 88217730; Fax: 86 10 68211420; http://www.casm.ac.cn.