Unconstrained freehand-written Chinese characters recognition by self-growing probabilistic decision-based neural networks

This paper presents the design of self-growing probabilistic decision-based neural networks (SPDNN) for the recognition of unconstrained freehand-written Chinese characters. In this research, the authors have developed: (1) an SPDNN based personal handwriting adaptive methodologies, (2) a two stage recognition structure: (a) a handprinted character recognizer, and (b) a personal adaptive freehand-written Chinese character recognizer, on a personal computer. For the unconstrained human handwriting, most of the reported handwriting recognition systems performed poorly (recognition rate falls between 40% and 50%). The proposed system shows significant improvement on the recognition rates through adaptive learning. The average recognition rates was raised from 44.09% to 82.2% in 5 learning cycles. And the performance could finally be increased up to 90.03% in 10 learning cycles