A fully-neural solution for online handwritten character recognition

The goal of this work is to provide a purely neuronal solution with no preprocessing for online handwritten character recognition. The idea consists of utilising the neurons enriched by a spatio-temporal (ST) coding developed in our laboratory. The coding, defined in the complex domain, is specially conceived for the processing of ST patterns. In this model, the stroke of a character generated by a digitizing tablet is presented in the form of a sequence of spikes corresponding to displacements of the stylus. The task of recognition comprises two steps. There is an initial layer of ST neurons which have the task of detecting certain primitives (lines) in the stroke of an alphabet. In the second step, we have a multilayer perceptron based on ST neurons, which recognizes the alphabet drawn from these primitives