Effective Handwritten Hangul Recognition Method Based on the Hierarchical Stroke Model Matching

This study defines three models based on the stroke for handwritten Hangul recognition. Those are trainable and not sensitive to variation which is frequently founded in handwritten Hangul. The first is stroke model which consists of 32 stroke models. It is a stochastic model of stroke which is fundamental of character. The second is grapheme model that is a stochastic model using composition of stroke models and the last is character model that is a stochastic model using relative locations between the grapheme models. This study also suggests a new stroke extraction method from a grapheme. This method does not need to define location of stroke, but it is effective in terms of numbers and kinds of stroke models extracted from graphemes of similar shape. The suggested models can be adapted to hierarchical bottom-up matching, that is the matching from stroke model to character model. As a result of experiment, we obtain 88.7% recognition rate of accuracy that is better than those of existing studies.