Learning to Generate Realistic Scene Chinese Character Images by Multitask Coupled GAN
Scene text recognition, is challenging due to the large appearance variances of the scene character. Recently, deep learning technique has shown its power for scene text recognition, but it requires enormous annotated data for training and it is time-consuming to manually obtain abundant data for all the categories of characters. This paper proposes a new architecture, called multitask coupled generative adversarial network (MtC-GAN), for scene Chinese character recognition (SCCR). The MtCGAN consists of coupled GAN networks for scene character style transfer and classifier networks trained by the style-transferred data generated by the coupled GAN. To make the generated data be realistic enough for SCCR, we train the multitask networks using a new loss function that combines the constrains of encoders, generators and classifiers simultaneously. Experiments show that the proposed MtC-GAN framework is general and flexible to improve the accuracy for SCCR.
Scene Chinese character recognition Generative adversarial networks Multitask training
Qingxiang Lin Lingyu Liang Yaoxiong Huang Lianwen Jin
School of Electronic and Information Engineering,South China University of Technology,Guangzhou,Chin School of Electronic and Information Engineering,South China University of Technology,Guangzhou,Chin
国际会议
广州
英文
41-51
2018-11-23（万方平台首次上网日期，不代表论文的发表时间）