Information-theoretic assessment of on-board near-lossless compression of hyperspectral data

Abstract

Abstract.A rate-distortion model to measure the impact of near-lossless compression of raw data, that is, compression with user-defined maximum absolute error, on the information available once the compressed data have been received and decompressed is proposed. Such a model requires the original uncompressed raw data and their measured noise variances. Advanced nearlossless methods are exploited only to measure the entropy of the datasets but are not required for on-board compression. In substance, the acquired raw data are regarded as a noisy realization of a noise-free spectral information source. The useful spectral information at the decoder is the mutual information between the unknown ideal source and the decoded source, which is affected by both instrument noise and compression-induced distortion. Experiments on simulated noisy images, in which the noise-free source and the noise realization are exactly known, show the trend of spectral information versus compression distortion, which in turn is related to the coded bit rate or equivalently to the compression ratio through the rate-distortion characteristic of the encoder used on satellite. Preliminary experiments on airborne visible infrared imaging spectrometer (AVIRIS) 2006 Yellowstone sequences match the trends of the simulations. The main conclusion that can be drawn is that the noisier the dataset, the lower the CR that can be tolerated, in order to save a prefixed amount of spectral information. © The Authors. Published by SPIE under a Creative Commons Attribution 3.0 Unported License. Distribution or reproduction of this work in whole or in part requires full attribution of the original publication, including its DOI.

Influence of the System MTF on the On-Board Lossless Compression of Hyperspectral Raw Data

“…In the case of DPCM-like strategies, the possibility to achieve a lossy compression with the additional feature of a peak error controlled by the user is also achievable, by inserting a quantization step in the prediction loop. This is the so-called near-lossless compression, which can be obtained by means of both causal and no-causal schemes [22,23]. Concerning the encoding stage, the arithmetic coding can reach the entropy limit [24], but it is not exploitable for an on-board implementation.…”