Statistical methods [for texture analysis]

Even though they do not appear to correspond to how the human visual system works, statistical methods are often used in trying to discriminate textures automatically. Correlations, conditional entropies and fractal dimensions are commonly computed. Often it is assumed that different parts of a texture are statistically independent, so that the texture can be characterized by probabilities for local patterns, as in a so-called Markov random field or generalized autoregressive moving average (ARMA) process.