Recognizing online handwritten Chinese characters using RNNs with new computing architectures

Abstract

Although online handwritten Chinese characters recognition has been explored for decades, it is still a challenging task to recognize handwritten Chinese characters accurately. In this paper, we propose an end-to-end recognizer for online handwritten Chinese characters recognition based on a new recurrent neural network (RNN). In the system, two new computing architectures are proposed based on traditional RNN system. One is the variance constraint, and the other is attention weight vector. The variance constraint is used to increase the representation ability of RNN, and the attention weight vector is used to describe the importance of hidden layer states at different time steps. Benefited from the two innovations, the recognition system obtains higher recognition accuracy with fewer parameters. Experiments are carried out on two handwritten Chinese character datasets, IAHCC-UCAS2016 dataset and ICDAR-2013 competition database. The experimental results show that the two innovations are effective, and the proposed end-to-end recognizer obtains better performance than the state-of-the-art methods.