INFORMATION-THEORETIC ASSESSMENT OF OPTICALREMOTE-SENSING IMAGERY

摘要

This work focuses on estimating the information conveyedrnto a user by multi-band remotely sensed opticalrndata, either multi-spectral or hyper-spectral. A trade-offrnexists between spatial and spectral resolution, due tornphysical constraints of sensors imaging with a prefixedrnSNR. Lossless data compression is exploited to measurernthe useful information content of the data. The bit-raternachieved by the reversible compression process takes intornaccount both the contribution of the "observation" noise,rnI.e. information regarded as statistical uncertainty, whosernrelevance is null to a user, and the intrinsic informationrnof hypothetically noise-free radiance data. An entropyrnmodel of the image source is defined and, once thernstandard deviation of the noise, assumed to be Gaussian,rnhas been preliminary measured, such a model is invertedrnto yield an estimate of the information content of thernnoise-free source from the code rate. Results of mutualrninformation assessment are reported and discussed onrnLandsat TM data and on AVIRIS data.