Handwritten Chinese Character Recognition: Effects of Shape Normalization and Feature Extraction

The technology of handwritten Chinese character recognition (HCCR) has seen significant advances in the last two decades owing to the effectiveness of many techniques, especially those for character shape normalization and feature extraction. This chapter reviews the major methods of normalization and feature extraction and evaluates their performance experimentally. The normalization methods include linear normalization, nonlinear normalization (NLN) based on line density equalization, moment normalization (MN), bi-moment normalization (BMN), modified centroid-boundary alignment (MCBA), and their pseudo-two-dimensional (pseudo 2D) extensions. As to feature extraction, I focus on some effective variations of direction features: chaincode feature, normalization-cooperated chaincode feature (NCCF), and gradient feature. The features are compared with various resolutions of direction and zoning, and are combined with various normalization methods. In experiments, the current methods have shown superior performance on handprinted characters, but are insufficient applied to unconstrained handwriting.