An invariant display strategy for hyperspectral imagery

Abstract

Remotely sensed data produced by hyperspectral imagers contains hundreds of contiguous narrow spectral bands at each spatial pixel. The substantial dimensionality and unique character of hyperspectral imagery requires display techniques that differ from those provided by traditional image analysis tools. This study investigated techniques enabling the display of hyperspectral images without the interference of inscene characteristics that lead to biased representations depending on the content of every image under analysis. Utilizing the Principal Components Analysis transformation it is possible to simplify the representation requirements while maintaining the info rmation contained in the scene. The introduction of an external eigenvector, containing few spectral characteristics, into the original scene data removes most of the spectral bias allowing for an accurate detection of the constituent elements. The subsequent shift of the resulting data to match the respective hue directions in the dataspace allows for image color fidelity based on the true composition of the image while all the environmental influence has been removed and the final outcome is readily perceived by the human vision.