Journal of Computer Applications ›› 2016, Vol. 36 ›› Issue (12): 3418-3422.DOI: 10.11772/j.issn.1001-9081.2016.12.3418

Previous Articles     Next Articles

Text segmentation based on superpixel fusion

ZHANG Kuang, ZHU Yuanping   

  1. College of Computer and Information Engineering, Tianjin Normal University, Tianjin 300387, China
  • Received:2016-05-16 Revised:2016-07-05 Online:2016-12-10 Published:2016-12-08
  • Supported by:
    This work is partially supported by the National Natural Science Foundation of China (61203259).


张矿, 朱远平   

  1. 天津师范大学 计算机与信息工程学院, 天津 300387
  • 通讯作者: 朱远平
  • 作者简介:张矿(1990-),男,河北石家庄人,硕士研究生,主要研究方向:图像处理、模式识别;朱远平(1978-),男,江西临川人,副教授,博士,主要研究方向:图像处理、模式识别。
  • 基金资助:

Abstract: Improving performance of text segmentation is an important problem in text recognition, which is disturbed by complex background and noises in text image. In order to solve the problem, a text segmentation method based on superpixel fusion was proposed. Firstly, the text image was binarized initially and text stroke width was estimated. Then, superpixel segmentation and superpixel fusion were completed in the images. Finally, the local consistence characteristic of the fused superpixel was taken to check the original binary image. The experimental results show that, compared with Maximally Stable Extremal Region (MSER) and Stroke based Superpixel Grouping (SSG), the segmentation precision of the proposed method is improved by 8.00 percentage points and 7.00 percentage points on KAIST Datebase, and the text recognition rate of the proposed method is improved by 5.33 percentage points and 4.88 percentage points on ICDAR2003 Datebase. The proposed method has strong ability of denoising.

Key words: text segmentation, superpixel, superpixel fusion, binarization

摘要: 提高复杂背景及噪声干扰文本图像的文本分割性能是文本识别研究中的重要问题和难点,为更好地解决这一难题,提出一种基于超像素融合的文本分割方法。首先对文本图像初始二值化,并估计文本笔画宽度;然后进行图像超像素分割并融合;最后利用超像素融合的局部相似性对初始二值化图像进行文本校验。实验结果表明,与最大稳定极值区域(MSER)及笔画超像素聚合(SSG)方法相比,所提方法在KAIST数据集上的分割精度分别提高了8.00个百分点和7.00个百分点,在ICDAR2003数据集上的文字识别率分别提高了5.33个百分点和4.88个百分点。所提方法具有较强的去噪能力。

关键词: 文本分割, 超像素, 超像素融合, 二值化

CLC Number: