Hyperspectral Data Processing: Algorithm Design and Analysis

Appendix

Algorithm Compendium

This appendix compiles many algorithms described in this book, most of which have been developed in the Remote Sensing Signal and Image Processing Laboratory (RSSIPL) at the University of Maryland, Baltimore County. In order to help readers implement these important algorithms, their MATLAB codes are also included for reference so that readers can write their own program codes without relying on software packages such as ENVI, ERDAS, etc. Each algorithm is described according to its functionality and its categorization and is then followed by its MATLAB codes.

A.1 Estimation of Virtual Dimensionality

The concept of virtual dimensionality (VD) was first coined in the book by Chang (2003a) and later published by Chang and Du (2004). It is defined as the number of spectrally distinct signatures present in the data and has received considerable interest in unsupervised hyperspectral data exploitation since it was introduced in 2003. Many approaches have been developed for estimating the value of VD. Nevertheless, the most popular algorithm to be used for this purpose is the one developed by Harsanyi et al. (1994a), whose idea was the origin of VD. All the details of VD along with techniques developed to estimate VD can be found in Chapter 5.

A.1.1 Harsanyi–Farrand–Chang Method

• Algorithm name: Harsanyi–Farrand–Chang (HFC) method
• Authors: J. C. Harsanyi and Chein-I Chang
• Category: preprocessing
• Designed criteria: eigenvlaues of sample correlation/covariance ...