论文信息 - Generalization of the EM algorithm for mixture density estimation

Generalization of the EM algorithm for mixture density estimation

The expectation-maximization (EM) algorithm is used for estimating mixture density parameters. This algorithm relies on the assumption that the number of component densities is given or known. This paper presents a preprocessing module to generalize the EM algorithm for the purpose of easing the assumption regarding the number of component densities. This module consists of a clustering algorithm, called multi-scale clustering, which allows an optimal number of component densities to be found by using scale-space theory. Examples are provided to (i) illustrate the improvement made by this generalization over the original EM algorithm and (ii) examine the performance of the developed algorithm in realistic situations.

Nasser Kehtarnavaz | Eiji Nakamura

[1] P. Green,et al. On Bayesian Analysis of Mixtures with an Unknown Number of Components (with discussion) , 1997 .

[2] M. J. Box. A New Method of Constrained Optimization and a Comparison With Other Methods , 1965, Comput. J..

[3] 中村栄治. Clustering based on scale-space theory , 1997 .

[4] D. Rubin,et al. Maximum likelihood from incomplete data via the EM - algorithm plus discussions on the paper , 1977 .

[5] Richard O. Duda,et al. Pattern classification and scene analysis , 1974, A Wiley-Interscience publication.

[6] A. F. Smith,et al. Statistical analysis of finite mixture distributions , 1986 .

[7] Heekuck Oh,et al. Neural Networks for Pattern Recognition , 1993, Adv. Comput..

[8] Simon Haykin,et al. Neural Networks: A Comprehensive Foundation , 1998 .

[9] R. Redner,et al. Mixture densities, maximum likelihood, and the EM algorithm , 1984 .

[10] Tony Lindeberg,et al. Scale-Space Theory in Computer Vision , 1993, Lecture Notes in Computer Science.