INFORMATION-THEORETIC ASSESSMENT OF LOSSY AND NEAR-LOSSLESS ON-BOARD HYPERSPECTRAL DATA COMPRESSION

This paper proposes a Rate-Distortion model to measure the impact of lossy and near-lossless compression of raw data on the information conveyed once the lossy/near-lossless decompressed raw data have been converted to radiance units. Input variables of the model are the original uncompressed raw data and their measured noise variances, according to a mixed photon + electronic noise model. Band-scaling gains and destriping coefficients for calibration, or equivalently radiance data obtained from the raw data, are also assumed to be available. The model makes use of advanced lossless/near lossless methods achieving the ultimate compression, regardless of their computational complexity. They are not to be implemented on board, but are used to measure the entropy of the data. Preliminary experiments on AVIRIS 2006 Yellowstone sequences show the trend of spectral information, i.e. information pertaining the ideal noise-free radiance source, without acquisition noise, versus either bit-rate/compression ratio or MAD/MSE distortion.