Evaluation of Quality and Information Content of CHRIS Hyper-Spectral Images

This work focuses on evaluating quality and estimating information of CHRIS hyperspectral images. Quality is assessed through the characterisation of the noise while information is estimated by means of an operative definition according to which the information content of a data set is given by the amount of information that cannot be predicted from the data that have already been acquired and, consequently, by the entropy of the derived prediction errors. A noise model is first verified and the parameters of the model are then estimated. Afterwards, lossless data compression is exploited to measure the entropy of the prediction errors through their bit-rate. The information content of the data is estimated by considering that the bit-rate achieved by the reversible compression process is due both to the contribution of the noise, whose relevance is null to a user, and to the hypothetically noise-free data. To perform the estimation of the information characterizing the noise-free data, an entropy-variance model is assumed for the ideal source. Once all the parameters of the model have been estimated, the entropy of the noise-free source is derived. Results are reported and discussed for hyper-spectral data sets acquired by CHRIS spectrometer. Information assessment is investigated before and after the radiometric correction process for evaluating the contribution of any effect introduced in the processed data. Different areas of the same image are then processed in order to assess the noise model. Eventually, the procedure is utilised to characterise data sets that have been acquired in different times in order to verify any potential operational change occurred in the instrument set-up or in the processing chain.