Image Data Compression Experiment Report. Volume 4: The Effect of Image Data Compression on a Parallelepiped Classifier

摘要

Irreversible data compression by cluster coding each line of raster scanned multispectral images is proposed for onboard reduction of telemetry data rates from European remote sensing satellites. Although changes to the data structure are small in terms of rms error, certain types of interpretation technique can be particularly sensitive. Simulation is one of the best ways of evaluating this interaction, and the results of a brief investigation of parallelepiped classification applied to data on an interactive facility is described. Results are satisfactory for field classification on both raw channels and principal components except where class distributions overlapped. In some cases, results on the reconstructed image data are better because of the smoothing effect of a reduction in intra-class variance.