Hyperspectral feature extraction using contourlet transform

In this paper, we explore hyperspectral feature extraction using the contourlet transform (CT), a promising multireolution analysis technique emerging in recent years. Hyperspectral imagery is first processed in the spectral domain with some decorrelation techniques. Then the nonsubsampled CT (NSCT) is applied in the spatial domain. The resulting NSCT coefficients are used as features for hyperspectral analysis. The spectral processing techniques being explored include one-dimensional discrete wavelet transform, principal component analysis, and band selection. The extracted features are tested in classification using support vector machine, which yield promising results.