Handbook of Biomedical Optics

Multi/Hyper-Spectral Imaging 151

Several PCA variants, such as independent component analysis

(ICA), selective PCA, Kernel-PCA (Yao and Tian 2003, Gu etal.

2006, Bayliss et al. 1997), or wavelet transform (Mallet et al. 1997,

Pu and Gong 2004), have been successfully used for spectral cube

data processing or as a preprocessing step for subsequent classi-

cation. Additionally, Filter (De Backer et al. 2005) and wrapper

methods (Bazi and Melgani 2006) as well as combinations (Mao

2004, Tan et al. 2008, Yu et al. 2002) have been employed to select

the best subset of spectral bands for optimal classication.

7.6.3 Spectral Classication and