TY - GEN
T1 - Triplet markov chains based- estimation of nonstationary latent variables hidden with independent noise
AU - Boudaren, Mohamed El Yazid
AU - Monfrini, Emmanuel
AU - Beghdad Bey, Kadda
AU - Habbouchi, Ahmed
AU - Pieczynski, Wojciech
N1 - Publisher Copyright:
© Springer International Publishing AG, part of Springer Nature 2018.
PY - 2018/1/1
Y1 - 2018/1/1
N2 - Estimation of hidden variables is among the most challenging tasks in statistical signal processing. In this context, hidden Markov chains have been extensively used due to their ability to recover hidden variables from observed ones even for large data. Such models fail, however, to handle nonstationary data when parameters are unknown. The aim of this paper is to show how the recent triplet Markov chains, strictly more general models exhibiting comparable computational cost, can be used to overcome this shortcoming in two different ways: (i) in a firmly Bayesian context by considering an additional Markov process to model the switches of the hidden variables; and, (ii) by introducing Dempster-Shafer theory to model the lack of precision of prior distributions. Moreover, we analyze both approaches and assess their performance through experiments conducted on sampled data and noised images.
AB - Estimation of hidden variables is among the most challenging tasks in statistical signal processing. In this context, hidden Markov chains have been extensively used due to their ability to recover hidden variables from observed ones even for large data. Such models fail, however, to handle nonstationary data when parameters are unknown. The aim of this paper is to show how the recent triplet Markov chains, strictly more general models exhibiting comparable computational cost, can be used to overcome this shortcoming in two different ways: (i) in a firmly Bayesian context by considering an additional Markov process to model the switches of the hidden variables; and, (ii) by introducing Dempster-Shafer theory to model the lack of precision of prior distributions. Moreover, we analyze both approaches and assess their performance through experiments conducted on sampled data and noised images.
KW - Data segmentation
KW - Hidden markov chains
KW - Nonstationary data
KW - Signal processing
KW - Triplet markov chains
U2 - 10.1007/978-3-319-93375-7_7
DO - 10.1007/978-3-319-93375-7_7
M3 - Conference contribution
AN - SCOPUS:85048977612
SN - 9783319933740
T3 - Lecture Notes in Business Information Processing
SP - 127
EP - 144
BT - Enterprise Information Systems - 19th International Conference, ICEIS 2017, Revised Selected Papers
A2 - Smialek, Michal
A2 - Hammoudi, Slimane
A2 - Camp, Olivier
A2 - Filipe, Joaquim
PB - Springer Verlag
T2 - 19th International Conference on Enterprise Information Systems, ICEIS 2017
Y2 - 26 April 2017 through 29 April 2017
ER -