A linear approach for two-dimensional, frequency domain, least square, signal and system modeling

A linear algorithm for two-dimensional (2-D) least square (LS) approximation in the frequency domain is presented. The algorithm is based on the equation error model. The approximation yields a 2-D rational function in the complex variables, or equivalently a 2-D autoregressive, moving-average (ARMA) process. The proposed two-dimensional, least square, frequency domain (2D-LS-FD) algorithm can efficiently represent 2-D signals or images. It is also capable of accurately modeling 2-D linear and shift invariant (LSI) stable systems, when the model has a sufficient order relative to the unknown and the identification noise is negligible. This paper will also discuss, with proofs, the important existence, uniqueness and convergence properties associated with this technique. Simulation examples for signal and system modeling are given to show the excellent performance of the algorithm. In addition, the successful application of the developed algorithm to image noise cancellation is also presented