Construction of stationary self-similar generalized fields by random wavelet expansion

Random wavelet expansion is introduced in the study of stationary self-similar generalized random fields. It is motivated by a model of natural images, in which 2D views of objects are randomly scaled and translated because the objects are randomly distributed in the 3D space. It is demonstrated that any stationary self-similar random field defined on the dual space of a Schwartz space of smooth rapidly decreasing functions has a random wavelet expansion representation. To explicitly construct stationary self-similar random fields, random wavelet expansion representations incorporating random functionals of the following three types are considered: (1) a multiple stochastic integral over the product domain of scale and translate, (2) an iterated one, first integrating over the scale domain, and (3) an iterated one, first integrating over the translate domain. We show that random wavelet expansion gives rise to a variety of stationary self-similar random fields, including such well-known processes as the linear fractional stable motions.