J4 ›› 2013, Vol. 40 ›› Issue (6): 187-192.doi: 10.3969/j.issn.1001-2400.2013.06.031

• 研究论文 • 上一篇    

一种图像中的文字区域检测新方法

李英1;田春娜1;颜建强1;庄怀宇2;李相威1   

  1. (1. 西安电子科技大学 电子工程学院,陕西 西安  710071;
    2. 中国移动通信集团 广东有限公司,广东 广州  510623)
  • 收稿日期:2012-11-07 出版日期:2013-12-20 发布日期:2014-01-10
  • 通讯作者: 李英
  • 作者简介:李英(1976-),女,西安电子科技大学博士研究生,E-mail: liying6@gd.chinamobile.com.
  • 基金资助:

    国家自然科学基金资助项目(61201291,61301192);高等学校博士学科点新教师基金资助项目(20090203120011)

New technique for text region location in images

LI Ying1;TIAN Chunna1;YAN Jianqiang1;ZHUANG Huaiyu2;LI Xiangwei1   

  1. (1. School of Electronic Engineering, Xidian Univ., Xi'an  710071, China;
    2. China Mobile Group Gangdong Co. Ltd., Guangzhou  510623, China)
  • Received:2012-11-07 Online:2013-12-20 Published:2014-01-10
  • Contact: LI Ying

摘要:

利用文字区域纹理的方向性,提出了一种结合Gabor纹理特征和神经网络分类器的图像文字区域检测算法.首先,通过不同方向和尺度下的Gabor特征来描述原始图像中文字区域的方向性纹理; 然后,将文字区域和非文字区域的Gabor特征输入到前馈神经网络,来训练文字区域分类器,训练好的分类器用于图像和视频中文字区域的检测.实验结果表明,该算法显著提高了文字区域检测算法的准确性和鲁棒性,对中、英文等多语种的文字区域检测均有较好效果.

关键词: 文字检测, Gabor特征, 神经网络

Abstract:

Text regions in images are hard to detect because of the complex background. Using the texture directions of the text region, we propose a new technique for text region location in images based on the Gabor character and neural networks classifier. First, the direction textures of original images are described by Gabor features with several directions and scales. Then the BP neural networks classifier is trained based on the Gabor features of text regions and the non-text regions, which can be used to detect the text regions. Experimental results illustrate that our algorithm greatly improves the accuracy and robustness than the others. And the proposed algorithm is suitable for locating not only the Chinese text but also the text in English and other languages.

Key words: text location, Gabor feature, neural networks

中图分类号: 

  • TP391.41
Baidu
map