Off-line Chinese handwriting recognition using multi-stage neural network architecture

In this paper, we propose a multi-stage neural network architecture (MNNA) which integrates several neural networks and various feature extraction approaches into a unique pattern recognition system. The general mechanism for designing the MNNA is presented. A three-stage fully connected feedforward neural networks system is designed for handwritten Chinese character recognition (HCCR). Different feature extraction methods are employed at each stage. Experiments show that the three-stage neural network based HCCR system achieved impressive performance and the preliminary results are very encouraging.