A HYBRID REVERSIBLE INTEGAR INTER-BAND DECORRELATION TECHNIQUE FOR LOSSLESS/LOSSY HYPERSPECTRAL IMAGE COMPRESSION

The placement of hyperspectral sensors on future Unmanned Air Vehicles is constrained by a data bottleneck caused by the limited downlink bandwidth. This proposal addresses this problem by developing a lossless compression algorithm that is optimized for hyperspectral imagery. Hyperspectral imagers generate 3-D data whose two axes correspond to spatial directions and the third to spectral response. The proposed solution exploits the latest results in lossless image compression by performing a reversible multiresolution decomposition of the data along the spatial axes. For extracting the redundancies in the spectral axis, one of three choices is proposed. The first is based on the principal component analysis, modified to ensure reversibility of the transformation. The second is based on cross-spectral prediction and the last on an integer multiresolution decomposition in the spectral direction. An experimental setup is proposed to pick the best candidate solution over a range of sample data. The transformed data is efficiently encoded by means of a contextual arithmetic encoder.Lossy compression is possible in a simple extension of the lossless technique. In addition the algorithm permits progressive transmission. The computational requirement of the algorithm is reasonable, and encoding the decoding complexities are nearly symmetric.