Off-line Handwritten Chinese Character Recognition Based On Stoke Structure Feature

Today, OCR technology for printed Chinese character has achieved the practical level. But, it is difficult to satisfy the demand of off-line handwritten Chinese character recognition. The main reason of this situation is the deformation of off-line handwritten Chinese characters. The single statistical feature extraction and the simple classification algorithm are not able to to adapt the changes of the handwritten Chinese characters'style. So, in this paper, we use template method in which combination distance of strokes as judgemental criterion to recognize off-line handwritten Chinese characters.The system consists of two stages, stroke structure feature extraction and matching. In the stage of stroke structure feature extraction, we get the Chinese characters'skeleton through thinning firstly, and then remove crossing points by measuring the degree of every pixel. The inflexion point can be got by max distance method, and then sub-stroke can be extracted by the crossing points and the inflexion point. In order to eliminate interference feature, the sub-strokes would be optimized. The sub-strokes which are linked together in original bitmap can be combined into strokes according to the similar matrix of sub-strokes. And the stroke structure feature would be constructed by using coordinates and the number of the strokes. In the stage of matching,using stroke structure features as the templates, the unknown character would match with the templates. The distances between the character and the templates would be calculated. And the character can be recognized by the above minimum distance. The experiments on off-line handwritten Chinese characters show that the method proposed in this paper is feasible.