Lossless compression of hyperspectral imagery with reversible integer transform

Hyperspectral imagery has very high spectral and spatial correlations. Since spectral information loss decreases the value of hyperspectral imagery for remote sensing applications, it is preferred to use lossless compression, although lossy compression can increase compression ratio. The Karhunen-Loeve Transform is theoretically the optimal transform to decorrelate hyperspectral data. However, since its transformed signal is real number, KLT is hardly applied in the field of lossless compression. The integer KLT (IKLT) based on triangular elementary reversible matrices (TERM) factorization of the transform matrix is perfectly reversible, and can be computed in place. A lossless decorrelating algorithm for hyperspectral imagery compression combining the integer KLT and the integer wavelet transform (IWT) is proposed. A complete-maximum pivoting is used to constructed integer approximation of the KLT and leads to only limited error and more computational efficiency. In addition, given its promising performance in still image compression, an integer wavelet transform is implemented by the lifting scheme and adopted as spatial decorrelating transform, which is also inversible. The experimental results with different coding schemes and hyperspectral imagery from different scenes show that our decorrelating method can significantly enhance compression ratio.