Handwritten Chinese character recognition method based on non-parametric dimensionality reduction

Dimensionality reduction is essential to Chinese character recognition. This paper proposes an optimal dimensionality reduction method based on the integration of geodesic paths and non-parametrical dimensionality reduction. In order to solve large scale pattern recognition, the paper presents simplification strategy for algorithms, which greatly speeds up training by employing non-parametrical dimensionality reduction algorithms optimized with geodesic distance on the premise that the recognition rate does not decline. It can be applied in both Chinese character recognition and digital recognition, which increases by 1.5 percentages in Chinese character recognition.