Abstract :
Hidden Markov chains, enabling one to recover the hidden process even for very large size, are widely used in various problems. On the one hand, it has been recently established that when the hidden chain is not stationary, the use of the theory of evidence is equivalent to consider a triplet Markov chain and can improve the efficiency of unsupervised segmentation. On the other hand, hidden semi-Markov chains can also be considered as particular triplet Markov chains. The aim of this paper is to use these two points simultaneously. Considering a non stationary hidden semi-Markov chain, we show that it is possible to consider two auxiliary random chains in such a way that unsupervised segmentation of non stationary hidden semi-Markov chains is workable
Keywords :
hidden Markov models; random processes; signal processing; auxiliary random chain; evidence theory; hidden semiMarkov chain; triplet Markov chain; unsupervised segmentation; Bayesian methods; Filtering; Hidden Markov models; Ice; Image sequence analysis; Kalman filters; Parameter estimation; Random variables; Roentgenium; Speech processing;
