Publication details

Multiple classifier fusion in probabilistic neural networks

Journal Article

Grim Jiří, Kittler J., Pudil Pavel, Somol Petr

serial: Pattern Analysis and Applications vol.5, 7 (2002), p. 221-233

research: CEZ:AV0Z1075907

project(s): GA402/01/0981, GA ČR

keywords: EM algorithm, information preserving transform, multiple classifier fusion

abstract (eng):

The main motivation of the present paper is to design a statistically well justified and biologically compatible neural network model and to suggest a theoretical interpretation of the high parallelism of biological neural networks. We consider a probabilistic approach to neural networks in the framework of statistical pattern recognition. The complete method based on EM algorithm has been applied to recognize unconstrained handwritten numerals from the database of the Concordia University Montreal.

Cosati: 12B, 09K