|
|
 |
| |
|
|
|
Knowledge-based Principal Component Analysis for Image Fusion |
|
|
|
PP: 223-230 |
|
|
|
Author(s) |
|
|
|
Jie-Lun Chiang,
|
|
|
|
Abstract |
|
|
| The purpose of image fusion is to integrate images with different resolution or from different sources in order to increase
information and reinforce identification and reliability in remote sensing application. The principal component analysis (PCA) approach
is a commonly used method for satellite image fusing. In the PCA fusion process, the first principal component (PC1) image is replaced
with a high resolution image (e.g., a Panchromatic (PAN) image of the SPOT4 satellite).When the histograms of PC1 and PAN images
are more similar, less spectral information is lost in the replacement process.
In this study, a knowledge-based principal component analysis (KBPCA) fusion is developed to improve the fusing results of PCA
approach. Before the replacement of PAN image, a prior landcover classification was done to gain the knowledge of the landcover of
study area. Principal component transform was then conducted on the individual data set of each landcover class. Since the spectrum
variation of each landcover class is smaller than that of the entire image, such pre-classification makes the PC1 of each class, have
less spectrum variation, compared to the PC1 of the entire image. Landcover information derived from pre-classification is used as
additional information to limit spectrum variation in each class for image fusion during the principal component transform. As a result
of fusion, a multi-spectrum high resolution new image can be produced by fusing multispectral and PAN images of SPOT4. The images
fused by the KBPCA method are of quality superior to those fused by the PCA method in terms of visual and statistical assessments. |
|
|
|
|
|
|
|
