Spectral reduction image processing techniques

Very high-resolution imagery (spatial or spectral) comes at a cost, large files for manipulation and the curse of dimensionality. To overcome these problems, it becomes necessary to reduce the spectral dimensionality of the data with a minimum loss of information. Several techniques will be examined in this paper. These include principal component analysis, singular value decomposition, greedy search, discrete wavelet transform greedy search, and self-organizing maps. The methods are applied to very high spectral resolution data, and results are compared.