1. Introduction
Comparable to multispectral image data, hyperspectral image has high spectral dimension, vast data and altitudinal interband redundancy which present a challenge to traditional data processing techniques. In many cases, it is unnecessary to process all the spectral bands of a hyperspectral image, since most materials have specific characteristics only at certain bands, which makes the remaining spectral bands somewhat redundant. Jimenez [1] pointed out that the hyperspectral data is centralized in low-dimensional space because of the high-dimensional space of hyperspectral image is relatively empty. Therefore, reducing the dimensionality of hyperspectral data without loosing important information about objects of interest is a very important issue for the remote sensing community. The dimension reduction of hyperspectral image is mainly divided into two categories, feature extraction and feature selection. Feature extraction transforms the original data from high dimension into low dimension with the most of the desired information content preserved. A number of methods have been developed to mitigate the effects of dimensionality on information extraction from hyperspectral data, such as principal component analysis (PCA) [2] and segmented principal components transform (SPCT) [3]. Because of not taking the class information of the input data into account, PCA may probably discard much useful information and weaken the recognition accuracy, especially when the number of sample points is very large [4].