Image processing in signal-dependent noise

Techniques for processing of images in signal-independent additive noise are well developed. However, in practice the noise is often dependent on the signal. Some attempts to take into account the dependence of the noise on the signal have been made: our approach has been to generalize the homomorphic transformation, which is a point transformation that transforms the output of a system into a space where the noise becomes independent of the signal. Once this has been accomplished, standard techniques such as Wiener filtering should be applicable with results that are predictable. A rigorous proof of the general homomorphic transformation is given, and its accuracy is discussed. Applications to speckle noise and to film-grain noise are presented.