신경망과 거부단을 이용한 필기체 숫자인식의 성능향상 = On Improving the Performance of the Handwritten Digit Recognition based on a Neural Network and Rejection Units

다국어 초록 (Multilingual Abstract)

The paper presents a method for combining a neural network and rejection units. Coventional neural network systems have such a drawback that they do not reject a garbage input pattern but always classify it as one of the predetermined outputs. This drawback is overcome by incorporating rejection units to the conventional MLP neural networks. A new concept has been proposed and applied to the design of the rejection units and the neural network. Several tests have been carried out to evaluate the performance of the proposed system in terms of the recognition rate, rejection rate and reliability. The simualtion results show that the proposed system performs better than the conventional ones.

목차 (Table of Contents)

• Ⅰ.序 論
• Ⅱ.필기체 숫자인식
• Ⅲ.신경망(MLP)의 구조
• Ⅲ-Ⅰ.오류 역전차 알고리즘
• Ⅳ.거부단(REJECTION UNIT)
• Ⅳ-Ⅰ.구조 및 인식방법
• Ⅳ-Ⅱ.거부단의 구현
• Ⅴ.신경망과 거부단의 병합
• Ⅵ.실험 및 검토
• Ⅶ.結 論