2008 
1223 
HYPERSPECTRAL RESOLUTION ENHANCEMENT USING HIGH-RESOLUTION 
IMAGERY WITH WAVELET PACKAGE ALGORITHM AND OPTIMAL INDEX 
PRINCIPLE 
D. G. Jun a , Z. Haifang b , Z. chaojie c 
a State Key Laboratory of Remote Sensing, Science, China, Beijing - topd@163.com 
b School of Computer Science, National University of Defence Technology, P. R. China, Changsha 
c Surveying and mapping group of Chengdu, P. R. China, Chengdu, Shuhan Street - zhu202@sohu.com 
Commission WG VII/6 
KEY WORDS: fusion, Hyper spectral, Processing, Imagery, Matching 
ABSTRACT: 
In this paper, aiming at image displaying and considering the characteristics of hyperspectral data and the requirements of practical 
application, a new hyperspectral image fusion method is proposed. This wavelet package image fusion approach is based on optimal 
index principle. The approach first selects optimal fusion bands of hyperspectral data by optimal index to construct synthesized low- 
resolution colour image and then fuses images by wavelet package algorithm which is based on multi-features in a region. The 
validity of the approach is testified by hyperspectral PHI data of shanghai area. Experiment shows that this approach can fuse 
hyperspectral data and high-resolution remote sensing data effectively and keep hyperspectral image’s spectral physical 
characteristics and shape, thus satisfies the requirements of practical applications and provides a better condition for further 
researches. 
1. INTRODUCTION 
The development and application of hyperspectral techniques 
request more advanced approaches to process the hyperspectral 
remote sensing data with large dimensions, and hyperspectral 
image fusion technique is a hot but fairly difficult research field. 
Within optical remote sensor systems, image’s spatial 
resolution is contradictive to its spectral resolution, which is 
that in case of known Signal-to-Noise, high spectral resolution 
(narrow spectral bands) is often obtained at the expense of 
spatial resolution. In order to solve this problem, hyperspectral 
satellite sensors such as LOUIS, ASTER, MODIS and so on 
are all equipped with high spatial resolution sensors, so that 
after fusion of spectral data and high-resolution spatial data, the 
spectral images spatial analytical characteristic is greatly 
improved and simultaneously its spectral physical 
characteristics and spectral information is efficiently kept 
(Zhang,2000). 
Presently there are many fusion approaches for multi-spectral 
remote sensing image and high-resolution panchromatic image. 
Because of limited data resources, hyperspectral image fusion 
methods are relatively less (Park,2000). Analyzing the 
characteristics of hyperspectral images, adopting the thought of 
spectral fusion and based on image spectral recovering 
approach, a new spatial remote sensing data fusion model 
(SFSR) which first applies histogram equalization to 
hyperspectral image and high-resolution panchromatic image 
and then fuse them on each spectral band separately is 
proposed (Zhang,2000). On the basis of adaptive spatial 
decomposing approach and based on local information entropy, 
a multi-scale wavelet fusion method to extract feature images is 
proposed (Zhang,2002). And by feature selection principle 
based on contrast sensitivity, Wilson proposed a hyperspectral 
image fusion method to avoid the data processing problem 
caused by massive data (Wilson, 1998). However, these multi 
scale fusion methods usually fuse images based on single 
feature and definitely have some limits (Zhang,2004). Moreover, 
wavelet transformation fusion method can decompose, fuse and 
reconstruct images iteratively, but these operations are only 
conducted on image’s low frequency part while its high 
frequency part (the image’s detailed part) is ignored. 
The combing of hyperspectral image and its spectrum and the 
characteristics of massive data make it difficult to fuse 
hyperspectral image while using common fusion methods. In 
this paper, aiming at image displaying and considering the 
characteristics of hyperspectral data and the requirements of 
practical application, a new hyperspectral image fusion method 
is proposed. This method can keep hyperspectral image’s 
spectral physical characteristics and shape, and thus satisfies the 
requirements of practical applications and provides a better 
condition for further researches. 
2. WAVELET PACKAGE FUSION METHOD FOR 
HYPERSPECTRAL IMAGE BASED ON OIF 
Wavelet transformation fusion method can decompose, fuse and 
reconstruct images iteratively, but these operations are only 
conducted on image’s low frequency part while its high 
frequency part (the image’s detailed part) is ignored. However, 
wavelet package can efficiently avoid this problem and further 
decompose image’s high frequency part on different scale, and 
thus obtain higher quality images. In practical applications, we 
are more concerned with improving the spatial resolution of 
fused image and ensure that its overall tone is consistent. 
Therefore, in this paper, an image fusing method based on OIF 
is proposed, the main idea of which is that: first select the 
optimal synthesizing bands of hyperspectral image by optimal 
index principle and obtain low-resolution synthesized colour 
image, then based on multi-features in a region, fuse it with