Two-dimensional modeling of image random field using artificial neural networks

The authors address the problem of 2-D linear modeling of an image random field using a neural network approach. A new learning scheme is developed using the recursive least squares (RLS) method which can be used to extract the model coefficients. Both 2-D autoregressive (AR) with nonsymmetric half-plane (NSHP) and noncausal region of support, and general 2-D autoregressive moving-average (ARMA) models with NSHP region of support are considered. The proposed scheme is inherently fast and ideally suited for real-time implementations. It does not need any prior statistical knowledge of the image process. Numerical results demonstrate the advantages of the proposed scheme over the conventional parameter estimation methods.<>