Extraction of spatial and spectral scene statistics for hyperspectral scene simulation

One of the major challenges in scene simulation is to realistically model the spatial and spectral statistics found in natural scenes. A method for the extraction of spectral and spatial scene statistics from hyperspectral data is presented. It is designed to work on atmospherically compensated data in any spectral region, although this paper will report on visible-shortwave infrared scene statistics derived from HYDICE data. Our approach is based on a physical description where the scene is composed of materials that in turn are described by a set of spectral endmembers. The statistical quantities that are extracted include endmember abundance means and variances and channel variances. To develop a realistic surface reflectance that conforms to the real clutter statistics, the correlations among the endmember abundances were estimated by deriving an abundance covariance matrix. The pixel-to-pixel correlations were modeled using a fast first-order autoregressive moving average texture generation technique. The generated texture reflectance surfaces are suitable for inclusion in scene generation models such as MCScene and DIRSIG.