LIKELIHOOD CALCULATION FOR A CLASS OF MULTISCALE STOCHASTIC-MODELS, WITH APPLICATION TO TEXTURE-DISCRIMINATION

A class of multiscale stochastic models based on scale-recursive dynam ics on trees has recently been introduced. Theoretical and experimenta l results have shown that these models provide an extremely rich frame work for representing both processes which are intrinsically multiscal e, e.g., 1/f processes, as well as 1-D Markov processes and 2-D Markov random fields. Moreover, efficient optimal estimation algorithms have been developed for these models by exploiting their scale-recursive s tructure. In this paper, we exploit this structure in order to develop a computationally efficient and parallelizable algorithm for likeliho od calculation. We illustrate one possible application to texture disc rimination and demonstrate that likelihood-based methods using our alg orithm achieve performance comparable to that of Gaussian Markov rando m field based techniques, which in general are prohibitively complex c omputationally.