Pairwise Markov Models and Hybrid Segmentation Approach

Abstract The article studies segmentation problem (also known as classification problem) with pairwise Markov models (PMMs). A PMM is a process where the observation and underlying state sequence form two-dimensional chain, it natural generalization of hidden model. To demonstrate richness class PMMs, we examine closer few examples rather different types PMMs: model for two related chains, that allows to an inhomogeneous chain conditional marginal homogeneous PMM, semi-Markov assumes one processes observed other not, estimate unobserved path given observations. standard estimators often used are so-called Viterbi (a maximum probability observations) or pointwise posteriori (PMAP) maximizes observations pointwise). Both these have their limitations, therefore derive formulas calculating hybrid which interpolate between PMAP path. We apply introduced algorithms studied in order properties methods, illustrate large variation behaviour methods PMMs. show method should always be chosen care by taking into account purpose modelling particular interest.