Abstract: The paper proposed a method of off-line handwritten character recognition based on hierarchical classification. The method simulates the produce of character recognition of human. When a man wants to recognize a character,he uses different strategy in different situation. If the character has a simple structure, he uses global features; if it looks similar with other character,he uses local features. We divided the classifier into macro layer and micro layer. The macro layer uses gradient feature to represent global feature, it simulates the simple target recognition produce; the micro layer uses principal curve feature to represent local feature, it simulates the similar form character recognition produce. We used confidence value to measure indeterminacy of the produce and result. We gave the definition of similar form character,and rules to telling them. The experimental results indicate that the method can effectively improve the recognition rate of off-line handwritten character, especially well in telling similar form character.

Key words: Hierarchical classification, Handwritten character recognition, Confidence value, Similar form character,Principal curve, Gradient