Modelling and assessment of signal-dependent noise for image de-noising

In this paper, a class of signal-dependent noise models that are encountered in image processing applications is considered. They are defined by the gamma exponent, which rules the dependence on the signal of the noise, and by the variance of a stationary zero-mean random process that generates the signal-dependent noise. An observation noise term, zero-mean, white and independent of the signal, is also considered to account for the electronics noise. A blind procedure is proposed for reliably measuring the model parameters directly from the noisy images irrespective of their texture content. Such methods are iteratively based on linear regression techniques applied to scatter-plots of local first-order statistics calculated on homogeneous areas and drawn with logarithmic scale. Adaptive LLMMSE filtering is embedded in the iteration stage to provide a rough estimate of noise-free image texture which allows to discriminate between homogeneous and textured pixels. Experiments on simulated noisy images demonstrate a high accuracy of noise assessment.