Two-dimensional autoregressive (2-D AR) model order estimation

Much research has been devoted to the area of one-dimensional autoregressive (1-D AR) and autoregressive moving average (ARMA) model order selection. The most well-known solutions for this problem are the Akaike information criterion (AIC), MDL, and the minimum eigenvalue (MEV) criteria. On the other hand, all works in the 2-D case have focused on the problem of parameter estimation. In this correspondence, we extend the previous criteria to the 2-D AR model order determination. The model is assumed causal, stable, and spatially invariant with p₁×p₂ quarter-plane (QP) support. Numerical examples are given to illustrate the effectiveness of each method