The effect of pixel correlations on class separability

The effect of correlated observations on class separability is examined for the case when pixel dependences can be modelled by the ARMA process. It is shown that depending on model parameters, class separability can be improved by decorrelating the observed variables using inverse filtering.