The effect of spectrally correlated noise on noise estimation methods for hyperspectral images

Accurately estimating the noise in a hyperspectral image is necessary for many applications, including noise whitening and dimension reduction, for example. Inaccuracies in the noise estimation may lead to incorrect results in the information extraction from the image. It is known that most real images contain noise that is correlated between spectral bands, and certain noise estimation methods are affected by this correlation. In this paper we will investigate three popular noise estimation techniques, viz. multiple regression, a residual technique, and a spatially based method. These will be tested on simulated data as well as real Cuprite images acquired by AVIRIS, SpecTIR and Hyperion.