Modified Residual Method for Estimation of Signal Dependent Noise in Hyperspectral Images

An accurate estimate of the underlying noise is required in many hyperspectral image processing algorithms. In this regard, the classic residual method [1] is one of the most commonly employed noise estimation methods for hyperspectral images. We recently proposed [2] an improvement to the classic residual method, however the improvement did not cater for the case of signal-dependent noise. This is the case for most of the modern hyperspectral imaging instruments where the magnitude of the underlying noise is dependent on the intensity of the signal [3 -5]. In this paper, we extend the framework proposed in [2] to the case of signal-dependent noise and propose a method to estimate parameters of the signal-dependent noise variance. It is then shown that the proposed method is able to accurately estimate the different noise parameters in artificial datasets and the method is also applied and analyzed in real datasets.