RECOGNITION OF HANDWRITTEN NUMERALS BY STRUCTURAL PROBABILISTIC NEURAL NETWORKS
Jǐŕı Grim, Pavel Pudil, Petr Somol · 2008
The well known ”beauty defect” of probabilistic neural networks is the biologically unnatural complete interconnection of neurons with all input variables. Despite of deep formal reasons of this undesirable property, it can be removed by a special subspace approach without leaving the exact framework of Bayesian decision-making. As shown in a recent paper the related structural optimization based on EM algorithm is controlled by an information criterion. In the present paper the method has been applied to recognize unconstrained handwritten numerals from the database of Concordia University, Montreal, Canada. The obtained recognition accuracy is comparable with the previously published results though it has been achieved without any preceding feature extraction.