Chinese Image Text Recognition with BLSTM-CTC: A Segmentation-Free Method

摘要：

　　This paper presents BLSTM-CTC(bidirectional LSTMConnectionist Temporal Classification),a novel scheme to tackle the Chinese image text recognition problem.Different from traditional methods that perform the recognition on the single character level,the input of BLSTM-CTC is an image text composed of a line of characters and the output is a recognized text sequence,where the recognition is carried out on the whole image text level.To train a neural network for this challenging task,we collect over 2 million news titles from which we generate over 1 million noisy image texts,covering almost the vast majority of common Chinese characters.With these training data,a RNN training procedure is conducted to learn the recognizer.We also carry out some adaptations on the neural network to make it suitable for real scenarios.Experiments on text images from 13 TV channels demonstrate the effectiveness of the proposed pipeline.The results all outperform those of a baseline system.

关键词： Chinese image text recognition BLSTM CTC Segmentation-free

作者: Chuanlei Zhai Zhineng Chen Jie Li Bo Xu

作者单位: Interactive Digital Media Technology Research Center,Institute of Automation,Chinese Academy of Sciences,Beijing 100190,China

会议类型: 国际会议

会议名称: 第七届全国模式识别学术会议(The 7th Chinese Conference on Pattern Recognition,CCPR2016)

会议地点: 成都

会议语种:英文

页码: 525-536

在线出版日期: 2016-11-03（万方平台首次上网日期，不代表论文的发表时间）