东北大学学报（自然科学版）

东北大学学报(自然科学版) ›› 2007, Vol. 28 ›› Issue (2): 165-168.DOI: -

• 论著 • 上一篇    下一篇

基于小波变换和SVM的文本区域定位

孙红星;赵楠楠;徐心和;   

  1. 东北大学信息科学与工程学院;辽宁科技大学电信学院;东北大学信息科学与工程学院 辽宁沈阳110004;辽宁鞍山114044;辽宁沈阳110004
  • 收稿日期:2013-06-24 修回日期:2013-06-24 出版日期:2007-02-15 发布日期:2013-06-24
  • 通讯作者: Sun, H.-X.
  • 作者简介:-
  • 基金资助:
    国家自然科学基金资助项目(60475036)

Text region localization using wavelet transform in combination with support vector machine

Sun, Hong-Xing (1); Zhao, Nan-Nan (2); Xu, Xin-He (1)   

  1. (1) School of Information Science and Engineering, Northeastern University, Shenyang 110004, China; (2) School of Electronics and Information Engineering, Liaoning University of Science and Technology, Anshan 114044, China
  • Received:2013-06-24 Revised:2013-06-24 Online:2007-02-15 Published:2013-06-24
  • Contact: Sun, H.-X.
  • About author:-
  • Supported by:
    -

RichHTML

0

PDF (PC)

805

摘要： 提出了一种基于小波变换和支持向量机(SVM)在数字图像中定位文本的方法.对图像进行小波变换,并在低频概貌和高频能量空间应用SVM提取文本的纹理特征,由SVM来决定当前的像素是文本类还是非文本类.因为SVM的分类结果可能存在噪声或错误,用形态学去噪和计算纹理能量的方法对SVM的分类结果进行后处理.小波变换和SVM的结合,不仅降低了输入空间样本的数量,而且利用了SVM适合于高维空间工作的特点,提高了文本提取的效率.实验结果表明,提出的方法可以快速有效地定位数字图像中的文本区域.

关键词: 文本检测, 纹理分析, 小波变换, 支持向量机(SVM), 形态学

Abstract: A method based on wavelet transform in combination with support vector machine (SVM) is proposed for localizing text region. After the image was decomposed by wavelet transform, SVM is used to extract the texture characteristics of text from the low-frequency vague image sub-space and high-frequency energy sub-space and determine what the current pixel belongs to, i.e., the text class or non-text class. Then, because there are possible noise and false text after SVM classification, the methods of morphological denoising and textural energy calculating are used to reprocess the results of SVM classification. The proposed method utilizes the characteristic that SVM suits high-dimension space so as to improve the efficiency of extracting text in addition to reducing the number of spatial sample inputs. Experimental results showed that the method can rapidly and effectively localize the text region in digital image.

中图分类号: