High-Accuracy Offline Handwritten Chinese Characters Recognition Using Convolutional Neural Network

Yi Jiang ； Yaohui Song

《電腦學刊》 31卷6期 (2020/12) Pp. 12-23

https://doi.org/10.3966/199115992020123106002

摘要

Offline HCCR (Handwritten Chinese Character Recognition) is an important research area of pattern recognition. Along with the rapid development of deep learning technology, offline HCCR has been got some breakthrough. Currently, most of the HCCR solutions are either computational inefficient or not good at accuracy. A new solution is essential to provide both high accuracy and computational efficiency. In this study, a convolutional neural network based high-accuracy offline HCCR system is presented. Compared with existing HCCR solutions, the proposed solution achieves higher recognition accuracy and recall rate but uses fewer convolutional layers and fewer parameters. Experiments on the CASIA-HWDB1.1 dataset show that the recognition accuracy, top-2 accuracy, and top-5 accuracy of the proposed solution is 97.06%, 99.25%, and 99.75%, respectively.

關鍵字

convolution neural network ； deep learning ； offline Chinese handwritten character recognition ； patter recognition