End-to-End Chinese Image Text Recognition with Attention Model

doi:10.1007/978-3-319-70090-8_19

CORC > 自动化研究所 > 中国科学院自动化研究所 > 数字内容技术与服务研究中心 > 听觉模型与认知计算

	End-to-End Chinese Image Text Recognition with Attention Model
	Fenfen Sheng1,2
	2017
会议日期	November 14-18, 2017
会议地点	Guangzhou, China
关键词	Chinese Images Text Recognition · End-to-end · Attention · Segmentation-free
DOI	10.1007/978-3-319-70090-8_19
英文摘要	This paper presents an attention-based model for end-to-end Chinese image text recognition. The proposed model includes an encoder and a decoder. For each input text image, the encoder part firstly combines deep convolutional layers with bidirectional Recurrent Neural Network to generate an ordered, high-level feature sequence, which could avoid the complicated text segmentation pre-processing. Then in the decoder, a recurrent network with attention mechanism is developed to generate text line output, enabling the model to selectively exploit image features from the encoder correspondingly. The whole segmentationfree model allows end-to-end training within a standard backpropagation algorithm. Extensive experiments demonstrate significant performance improvements comparing to baseline systems. Furthermore, qualitative analysis reveals that the proposed model could learn the alignment between input and output in accordance with the intuition.
学科主题	Computer Vision
内容类型	会议论文
源URL	[http://ir.ia.ac.cn/handle/173211/19661]
专题	数字内容技术与服务研究中心_听觉模型与认知计算
作者单位	1.University of Chinese Academy of SciencesBeijingChina 2.Institute of AutomationChinese Academy of SciencesBeijingChina
推荐引用方式 GB/T 7714	Fenfen Sheng. End-to-End Chinese Image Text Recognition with Attention Model[C]. 见:. Guangzhou, China. November 14-18, 2017.