Feature preserving compression for hyperspectral remote sensing images

This paper proposes a dimensionality reduction and compression method of hyperspectral images based on independent component analysis (ICA) and mixed contourlet and wavelet transform (MCWT) for hyperspectral image analysis. At first, the spectral correlations of hyperspectral images are removed using ICA and dimensionality reduction is accomplished. Then, dimensionality reduction images are compressed by MCWT and set partitioning in hierarchical trees (SPIHT)-like coder. The experimental results by using 64 band hyperspectral data show that the proposed compression method preserves more spacial detail information and spectral features of hyperspectral images and achieves higher peak signal-to-noise ratio at high compression ratio than the compression method based on principal component analysis and wavelet transform.