A MDRNN-SVM hybrid model for cursive offline handwriting recognition

Abstract

This paper presents a recurrent neural networks applied to handwriting character recognition. The method Multi-dimensional Recurrent Neural Network is evaluated against classical techniques. To improve the model performance we propose the use of specialized Support Vector Machine combined whit the original Multi-dimensional Recurrent Neural Network in cases of confusion letters. The experiments were performed in the C-Cube database and compared with different classifiers. The hierarchical combination presented promising results. © 2012 Springer-Verlag.