Off-line Handwritten Chinese Character Recognition Based On Mathematical Morphology

With the increasing information communication and wide computer application, thedemand to process and recognize handwritten Chinese characters already becomes more andmore important. In many fields such as post, revenue, finance, insurance, office affairs and evendaily work, handwritten Chinese characters are required to be input into computer foraccelerating linguistic information communication. At present, this work mainly depends onmanual input. Obviously, inefficient manual input can not meet the need of high-speedautomatic Chinese characters recognition. Therefore, automatic handwritten Chinese charactersrecognition becomes an urgent problem.Off-line handwritten Chinese character recognition (HCCR) is a formidable task in thepattern recognition field. This research not only involves all of the typical problems in patternrecognition, but also has broad applications. Feature extraction is of great significance forhandwritten Chinese character recognition. The stability and efficiency of the feature havecrucial influence on the recognition system. This thesis aims at research of feature extraction,and besides, an experimental off-line handwritten Chinese character recognition system isdesigned.From the two aspects of statistical and structural features, this paper proposes a featureextraction method based on stroke decomposed by mathematical morphology and elasticmeshes. This method doesn’t need the thinning preprocess. By erosion and dilation operationsof the mathematical morphology and using different adaptive structure elements, the charactersare decomposed into strokes. Then the elastic meshes are applied to extract the directionalfeature and the crossing line number feature. Finally, combine the two eigenvectors, which havebeen normalized, to form complex vector, after generalized Karhunen-Loeve transform is usedto decrease the dimensions of these two feature vectors.The experiment results indicate the effectiveness of the novel method. And the improvingdirection is brought forward. It provides a good foundation for research and practical applicationlater.