Aspects of signal-dependent noise characterization

Abstract

An automated method is presented for analyzing signal-dependent noise. Signal-dependent noise is present in many types of data-acquisition processes and has been investigated by other researchers with various methods. Regardless of the noise analysis methods, often the starting point is based on a particular signal-dependent noise model, which also forms the basis for our study. The approach determines whether the estimated noise variance is dependent on the signal by approximating the functional relation within the constraints of the assumed signal-noise model. The method relies on the Fourier attributes of the signal and noise and uses the wavelet expansion for separating these components. The technique does not rely on the underlying noise and signal probability distributions. Two-dimensional simulations as well as mammography data are used to illustrate the merits of the approach.