Reversible Integer Principal Component Transform for Hyperspectral Imagery Lossless Compression

Hyperspectral imagery has very high spectral and spatial correlations. The principal component transform (PCT) is theoretically the optimal transform to decorrelate hyperspectral data. However, since its transformed signal is real number, PCT is hardly applied in the field of lossless compression. The integer PCT based on factorization of the transform matrix in triangular elementary reversible matrices (TERM) is computed in place and perfectly reversible. We investigate a compression technique combining the modified integer PCT and the integer wavelet transform (IWT). The modified integer PCT is constructed by complete-maximum pivoting that leads to only limited error and improved computational efficiency. Differing from the conventional PCT, the modified integer PCT is a segmented version for saving computation resources in real-time circumstances. Finally, an integer wavelet transform implemented by the lifting scheme is adopted as the spatial decorrelating transform, given its promising performance in still image compression. The experimental results of different coders and imagery from different scenes show that the proposed compression technique enhances compression ratio significantly.