基于CTPN的网店工商信息提取系统的研究和实现  

Research and Implementation of Online Business Information Extraction System Based on CTPN

作　　者：靳振伟 JIN Zhenwei(NingxiaUniversity,School oflnformation Engineering,Yinchuan 750021,China)

机构地区：[1]宁夏大学信息工程学院,宁夏银川750021

出　　处：《现代信息科技》2018年第11期27-28,共2页Modern Information Technology

摘　　要：本文提到的系统首先对图片进行预处理,消除水印等图片噪声,将图像二值化。CTPN利用同一文本线的字符可以互用上下文的特点,降低了检测难度。DenseNet在ResNet的基础上进行改进,减少了参数,改善了神经网络的性能。使用CTPN进行文本检测,使用DenseNet和CTC进行文本识别,获得了较高的准确率。The system mentioned in this paper firstly preprocesses the image,eliminates the image noise such as watermark,and binarizes the image. CTPN can reduce the difficulty of detection by using the characters of the same text line. DenseNet is improved on the basis of ResNet,which reduces parameters and improves the performance of neural network. CTPN is used for text detection DenseNet and CTC are used for text recognition,and higher accuracy is obtained.

关 键 词：文本检测 文本识别 CTPN 

分 类 号：TP393.09[自动化与计算机技术—计算机应用技术] TP311.13[自动化与计算机技术—计算机科学与技术]