Multivariate structural Bernoulli mixtures for recognition of handwritten numerals
J. Grim, Pavel Pudil, Petr Somol · 2002
The structural optimization of a probabilistic neural network can be included into an expectation maximisation (EM) algorithm by introducing a special type of mixtures. The method has been applied to recognize unconstrained handwritten numerals from the database of Concordia University in Montreal. We discuss the possibility of a proper initialization of the EM algorithm for estimating the class-conditional multivariate Bernoulli mixtures.