On The Spectral Correlation Structure of Hyperspectral Imaging Data

Spectral correlation, as quantified by the elements of the covariance matrix, plays a prominent role in the development of optimum statistical algorithms for hyperspectral data exploitation. Indeed, the most useful statistical models for hyperspectral image modeling, namely the multivariate normal distribution and the multivariate t-distribution, are parameterized by the spectral covariance matrix. The inverse of the covariance matrix, however, also has important interpretations. In this paper, we discuss the properties connected with the inverse covariance matrix and we describe their use in hyperspectral data analysis.