An identification method for texture AR-2D modelling based on auto- and partial correlation measures

This paper deals with the identification of the model order for a bidimensional autoregressive (AR-2D) texture model. It means the automatic choice of the number of neighbours in the prediction set of the model and their spatial position. The method, called mixed correlation method is based on partial and autocorrelation measures and fastly and efficiently allows to find an adapted model for all microtextures. In a textured samples classification procedure, these adapted models improve the percentage of good classification in comparison with a classical approach consisting in taking the same prediction set for all the textures.