3-Dimensional Signal Analysis of Multichannel Spectrometric Imagery

Hyperspectral, multispectral and multitemporal satellite images are considered as 3D signals and processed using 3D signal techniques. The third dimension is given by the spectral/temporal channels and the idea to perform a 3D processing is an attempt to take advantage of the correlation existing between these channels in order to achieve higher data compression rates. Two methods are investigated and compared: a 3D linear predictor based on a low order Markov model and a 3D wavelet decomposition procedure. Although the compression rate depends highly on the structure of the data, the 3D algorithms perform always better than the 2D ones. This justifies the use of 3D signal processing techniques in remote sensing applications involving a large number of correlated spectral or temporal channels.