Department of Computer and Information Engineering, Henan Key Laboratory of Oracle Bone Inscriptions Information Processing and Collaborative Innovation Center of International Dissemination of Chinese Language of Henan Province, Anyang Normal University, Anyang, China.

---

Abstract—In this paper, we investigate the recognition of oracle bone inscriptions (OBIs). By making use of the powerful ability of convolutional neural network (CNN) to describe image features, we design a CNN-based method to recognize OBIs. Our key insight is to build a CNN model for the recognition of OBI characters, which employs only image filters with 3×3 receptive field. Some commonly used techniques such as rectified linear unit (ReLU) and data augmentation have been employed to guarantee the accuracy of our method. Besides, the adaptive moment estimation (Adam) algorithm is used for training because of its straightforward implementation, computational efficiency and good empirical performance. In this paper, an OBI dataset with 44868 OBI images including 5491 different OBIs is established to testify the performance of the proposed method. Experimental results have shown the superiority to the state-of-the-art method.

Index Terms—Character recognition, deep convolutional neural network, oracle-bone inscription, oracle-bone inscription recognition, small kernel size.

[PDF]

Cite: Guoying Liu, Feng Gao, "Oracle-Bone Inscription Recognition Based on Deep Convolutional Neural Network," Journal of Computers vol. 13, no. 12, pp. 1442-1450 , 2018.