A light and discriminative deep networks for off-line handwritten Chinese character recognition
Shimeng Yang, Fudong Nian, Teng Li · 2017
A practical method to improve the performance of off-line Handwritten Chinese Character Recognition (HCCR) was proposed in this paper. The center loss was used in face verification task to optimize intra-class distance. With the joint supervision of softmax loss and center loss, A light convolutional neural networks (CNNs) framework was trained for off-line HCCR which could optimize inter and intra difference simultaneously. Extensive experimental results on pubic dataset demonstrate that the performance of HCCR is improved significantly.