Characters Recognition of Korean Historical Document Base on Data Augmentation
Chun-Han Xue, Xiaofeng Jin · 2020 5th International Conference on Mechanical, Control and Computer Engineering (ICMCCE) · 2020
Character recognition of historical document is one of the most important basic tasks in the digitization of historical document. This paper is aimed at the few-shot learning problem of Korean ancient character recognition, a data enhancement method that combines traditional data and a conditional deep convolution generation confrontation network is proposed to obtain expanded samples. Then analyzed the performance of the Lenet@5 and Lenet@8 network models in Korean ancient character recognition. The experimental results show that the expanded sample significantly enriches the experimental data, and the improved Lenet@8 network model is better than Lenet@5, which can better acquire image features and greatly improve the classification accuracy. The proposed method can solve the problem of Recognition of Text-Image Characters of Korean Historical Document. Keywords-Character recognition of Korean historical document; Few-shot learning; Data augmentation; Generative Adversarial Networks; Convolutional neural network.