INVESTIGADORES
DI PERSIA Leandro Ezequiel
congresos y reuniones científicas
Título:
Signal denoising with hidden Markov models using hidden Markov trees as observation densities
Autor/es:
DIEGO MILONE; LEANDRO DI PERSIA; DIEGO TOMASSI
Lugar:
Cancun
Reunión:
Workshop; IEEE Workshop on Machine Learning for Signal Processing; 2008
Institución organizadora:
IEEE
Resumen:
Wavelet-domain hidden Markov models have been found successful in exploiting statistical dependencies between wa-velet coefficients for signal denoising. However, these mo-dels typically deal with fixed-length sequences and are notsuitable neither for very long nor for real-time signals. Inthis paper, we propose a novel denoising method based on aMarkovian model for signals analyzed on a short-term basis.The architecture is composed of a hidden Markov model inwhich the observation probabilities are provided by hiddenMarkov trees. Long-term dependencies are captured in theexternal model which, in each internal state, deals with thelocal dynamics in the time-scale plane. Model-based denoi-sing is carried out by an empirical Wiener filter applied tothe sequence of frames in the wavelet domain. Experimen-tal results with standard test signals show important reduc-tions of the mean-squared errors.
