Neural versus syntactic recognition of handwritten numerals
Luciana Veloso, J.M. de Carvalho · 1999
This work concerns the analysis, implementation and evaluation of three different methods for handwritten numerical character recognition. The first approach uses a classifier based on syntactical analysis by decision tree. The other two methods consist of: (a) a conventional feedforward multilayer neural network; and (b) a recurrent neural network, for which the elements of the output layer are all interconnected. The CENPARMI database was utilized for evaluation of the systems.