Estimation of images modeled by a two-dimensional separable autoegressive process

This note considers the estimation of two-dimensional images that may be modeled by a separable autoregressive process. We first derive a one-dimensional vector stochastic model with multiple delays for images; the one-dimensional vector model is further decomposed into a set of nearly independent equations using the matrix factorization theorem and the orthogonal sine transform [10]. Then, applying the kalman filter, the approximate feasible estimation algorithm is obtained.