A new cumulant-based lattice algorithm for adaptive identification of non-Gaussian bidimensional AR processes

This paper deals with bidimensional modeling using cumulants. Two main contributions are provided. Firstly, we show that the cumulant matrix involved in the 2D normal equations can be seen as the cumulant matrix of a multichannel process. Secondly, we use this equivalence to derive a 2D adaptive lattice algorithm based on cumulants, which can be seen as an extension of Swami´s method (1988) to the 2D case. The proposed algorithm has the advantage of yielding non-biased estimates of the 2D-AR parameters even when the image under study is disturbed by Gaussian noise