자소 분리에 의한 한글 인식에 관한 연구 = A Study on a Recognition of Hangul Characters by a Segmentation of Letter

다국어 초록 (Multilingual Abstract)

It is difficult for computer to recognize Hangul characters becauae they are composed of several letters and the letters are joined together in some characters. To recognize the characters, characters are classified as six types known as 6-type of Hangul, and letters of the classified characters are segmented according to 6-type of character, and the segmented letters are recognized using a neural network. The neural network for type classification and character recognition is constructed as multi-layer. Inputs of the neural network are 4-direction vectors which are extracted from the mesh of the normalized character. The neural network is learned by the training rule of back propagation and descending epsilon algorithm to increase the learning speed and classification rate in the character classification and is learned by the back propagation algorithm in the character recognition.
The type of character is classified completely in 1-type and 2-type and classification errors occur in the other four types. The most of recognition errors occur in 5-type and 6-type. It has been shown that classification rate is 96.5% and recognition rate is 93.5%.