225
A NOVEL IMAGE FUSION METHOD BASED ON 2DPCA IN REMOTE SENSING
Xue-ming Wu*, Wu-nian Yang
RS and GIS Institute, Chengdu University of Technology, Chengdu 610059, China,
nihaowxm@ 163.com, ywn@cdut.edu.cn
KEYWORDS: Image Fusion, Remote Sensing, PCA, 2DPCA
ABSTRACT:
PCA-Based image fusion algorithm, which is widely applied in remote sensing, is a typical technique in image fusion. It is easy to
implement and has good performance. However, PCA has two weaknesses. One is that an image must be transformed into a 1-D
vector when PCA is applied to the image and can not utilize its structural information. The other is that the spectral information is
badly lost in the fused image, although the spatial resolution is apparently improved. To avoid these disadvantages of PCA, this paper
shows a novel image fusion method based on two-dimensional PCA (2DPCA), which is directly applied to the image matrices
instead of 1D vector. Therefore, the structural information of the image is effectively utilized. Furthermore, with this new technique,
not only the spatial resolution of the fused image is greatly improved, but also the spectral information is well preserved. The results
of our experiment show the better flexibility of this method proposed in contrast to the traditional method.
1. INTRODUTION
The objective of image fusion is to use many images of the
same scene provided by different sensors in order to provide a
complete understanding of the scene, not only in terms of
position and geometry, but more importantly, in terms of
semantic interpretation (Henri, Maiter, et al., 1997; Towinn,
Taxt, 1998). Image fusion technique has been widely applied to
image processing, remote sensing, computer vision, and
military (QIN Zheng, et al., 2007). PCA is one of the statistical
methods for dimensionality reduction and correlation
elimination. It has been widely used in remotely sensed image
processing, including image coding, data compression, image
enhancement, multitemporal dimensionality and image fusion,
and so on (C. Pohl, J. L. Van Genderen, 1998). PCA-based
algorithm on remotely sensed image is a typical technique in
image fusion. PCA was first applied to image fusion in remote
sensing by Chevas P S, et al. (Chevas P S, et al., 1984; Chevas P
S, 1986; Chevas P S, et al., 1991) since 1980s. They integrated
multisensor and multiresolution images of the same scene, for
example, merging the Landsat TM images and SPOT
panchromatic images, to get new images about the scene. Since
then, further research on the technique was done by Cliche, G.
Bonn (Cliche, G Bonn, 1986), Ashbindu Singh (Ashbindu
Singh, 1985), James R. Carr (James R. Carr, 1998), and C. Pohl,
J. L. Van Genderen et al. (C. Pohl, J. L. Van Genderen, 1998).
This makes the PCA-based algorithm became an important
image fusion method in remote sensing. A typical application
of PCA-based algorithm is to integration the multispetral
images with low spatial resolution and the panchromatic
images with high spatial resolution. It is easy to implement and
has good performance.
However, PCA has two weaknesses. One is that an image must
be transformed into a 1-D vector when PCA is applied to the
image and can not utilize its structural information. The other is
that the spectral information is badly lost in the fused image,
although the spatial resolution is apparently improved.
2DPCA, which is first proposed in 2004 by Jian Yang, et al.
(Jian Yang, et al., 2004), is a technique on image analysis.
2DPCA is directly applied to the image matrices instead of ID
vector, and then the eigenvalues and the eigenvectors are
evaluated. In contrast to PCA, 2DPCA has better performance
in feature extraction and face recognition; especially it takes
less time and improves recognition rate (Jian Yang, et al., 2004).
Therefore, more attention has been paid on it (D.S. Guru, P.
Nagabhushan and B.H. Shekar, 2007; Hui KONG, Xu-chun LI,
et al., 2005; Hui KONG, Lei WANG, et al., 2005).
Now that 2DPCA has upstanding performance in feature
extraction and face recognition, can it be introduced in image
fusion? If it can, the performance of 2DPCA is better or worse
than PCA? This is the objective of this paper. After studying the
characteristic of PCA-based algorithm and 2DPCA, we
proposed a novel image fusion algorithm based on 2DPCA,
which is an entirely different algorithm in contrast to
PCA-based method. In the novel technique, not only 2DPCA is
introduced into remotely sensed image fusion, but also it has
better performance than PCA-based method. It is of importance
that the characteristic of 2DPCA-based technique overcome the
weakness of that of PCA-based. With this new technique, not
only the spatial resolution of the fused image is greatly
improved, but also the spectral information is well preserved.
The remainder of this paper is organized as follows. In Section
2, PCA-based image fusion algorithm on remotely sensed
images is described. The idea of 2DPCA and the proposed
2DPCA-based method are shown in Section 3. In Section 4, the
experimental results are presented for PCA-based and
2DPCA-based algorithms to demonstrate the effectiveness of
2DPCA-based method. Finally, conclusions are summarized in
Section 5.
2. PCA-BASED ALGORITHM
PCA-based image fusion algorithm on remotely sensed images
can be summarized to three steps as follows.
* Corresponding author.
