A Recognition Method of the Similarity Character for Uchen Script Tibetan Historical Document Based on DNN

Autor: Xiaojuan Wang, Yuehui Han, Weilan Wang, Yiqun Wang, Zhanjun Hao, Zhenjiang Li
Rok vydání: 2018
Předmět:
Zdroj: Pattern Recognition and Computer Vision ISBN: 9783030033378
PRCV (3)
DOI: 10.1007/978-3-030-03338-5_5
Popis: In order to improve the similarity character recognition of Tibetan historical document, this paper applied the Depth Neural Network (DNN) to similar characters recognition of Tibetan historical document, and proposed a recognition method of the similarity character for Uchen Script Tibetan based on deep learning. The effective feature learning and recognition are automatically carried out by DNN. We also introduced a sample labeling method of Tibetan historical document of Uchen Script using unsupervised clustering and constructing sample sets of the similar characters. Compared with the traditional methods such as Support Vector Machine (SVM) and Naive Bayes Classifier (NBC) based on gradient features through simulation experiment, our method can achieve better performance. The proposed method can learn feature effectively and avoid the disadvantages of manual feature selection and extraction, and it can improve recognition rate greatly. With the increasing of training samples, the recognition rate was improved more significantly. The experimental results show that the proposed method used for similar characters of Tibetan historical document Uchen Script recognition, higher recognition rate can be obtained.
Databáze: OpenAIRE