Application of EOF´s to multispectral imagery: data compression and noise detection for AVIRIS

Investigates the first stage of a two stage approach to data compression for multispectral imagery. The first stage is to compress the data spectrally using empirical orthogonal functions (EOFs). In the second stage, each EOF image is further compressed using standard techniques, such as transform encoding. The characteristics of EOFs make them ideal for spectral compression. The EOFs form the orthogonal basis in the data space which provides the most economical data representation. Furthermore, and perhaps of more interest, EOFs are effective noise filters. In the authors experiments with the 224 channel AVIRIS data, lossy compression ratios of order 50:1 are attained by the EOF representation under the condition that the residual rms error be smaller than the independently measured instrument noise