Likelihood calculation for a class of multiscale stochastic models

In this paper, we present an efficient likelihood calculation algorithm for a class of multiscale models. Our development exploits the scale-recursive structure of these models thereby leading to a computationally efficient and highly parallelizable algorithm. We illustrate one possible application of the algorithm to texture discrimination and show that likelihood-based methods using our algorithm perform have substantially better probability of error statistics than well-known least-squares methods, and achieve virtually the same performance as truly optimal techniques, which are prohibitively complex computationally