Unsupervised segmentation of switching pairwise Markov chains

Pairwise Markov chains (PMC) have now shown their supremacy over hidden Markov chains (HMC) in unsupervised data segmentation since they allow one to deal with more complex processes structures. HMCs are particular cases of PMCs and these latter provide a gain in restoration accuracy within comparable computational complexity. On the other hand, the recent triplet Markov chains (TMC) have successfully substituted for classical HMCs to model data with some irregularities that these latter cannot handle. In fact, they provide an elegant formalism through the introduction of a third underlying process that permits to consider, for instance, regime switches or semi- Markovianity of the hidden process. The aim of this paper is to generalize the switching HMC to switching PMC. To validate the proposed model, we choose non stationary image segmentation as illustrative application field. Experimental results of synthetic and real images segmentation are provided.