计算机应用 ›› 2011, Vol. 31 ›› Issue (10): 2697-2701.DOI: 10.3724/SP.J.1087.2011.02697

• 图形图像技术 • 上一篇    下一篇

适用于小样本问题的有监督边界检测方法

高梁1,廖志武2,刘晓云1,陈武凡1,3   

  1. 1.电子科技大学 自动化工程学院,成都 611731
    2.四川师范大学 计算机科学学院,成都 610101
    3.南方医科大学 生物医学工程学院,广州 510515
  • 收稿日期:2011-04-12 修回日期:2011-06-07 发布日期:2011-10-11 出版日期:2011-10-01
  • 通讯作者: 高梁
  • 作者简介:高梁(1980-),女,湖南汨罗人,博士研究生,主要研究方向:模式识别、图像处理;廖志武(1969-),女,湖南新田人,副教授,主要研究方向:图像处理、模式识别;刘晓云(1963-),女,四川南溪人,副教授,主要研究方向:图像处理、模式识别;陈武凡(1949-),男,湖南汨罗人,教授,博士生导师,主要研究方向:模式识别、医学图像处理。
  • 基金资助:

    国家973计划项目(2010CB732501)

Supervised boundary detection for small sample problem

GAO Liang1, LIAO Zhi-wu2, LIU Xiao-yun1, CHEN Wu-fan1,3   

  1. 1.School of Automation Engineering, University of Electronic Science and Technology of China, Chengdu Sichuan 611731,China
    2.College of Computer Science, Sichuan Normal University, Chengdu Sichuan 610101, China
    3.School of Biomedical Engineering, Southern Medical University, Guangzhou Guangdong 510515, China
  • Received:2011-04-12 Revised:2011-06-07 Online:2011-10-11 Published:2011-10-01
  • Contact: Liang GAO

摘要: 针对自然图像纹理复杂的特点,提出了一种多种信息融合的有监督边界检测方法。首先,该方法在小样本的情况下,通过快速生成纹理基元特征来引入纹理信息;然后,根据图像中每个像素邻域内的灰度分布和纹理基元分布的差异来计算灰度梯度和纹理梯度,并在此基础上构造出二维的梯度特征向量;接着,用有监督的分类器进行分类,自适应地检测出初始的边缘点;最后,设计一个边界定位函数确定最终的边缘点,实现边界检测。实验结果表明,该算法运算速度较快,所检测的边界效果好。

关键词: 小样本问题, 边界检测, 纹理基元, 监督学习, 分类器

Abstract: For natural images of complex texture, a supervised boundary detection method using the multi-information fusion was proposed. The texture information was introduced by quickly generating texton feature in the case of small sample. Intensity and texture gradients were further computed according to the differences of intensity and texton distributions within a pixel's neighborhood. In this way, a two-dimensional gradient feature vector was constructed, and a supervised classifier was used to adaptively detect original edge pixels. Finally, a boundary localization function was designed to determine the final edge pixels. The experimental results have demonstrated that the proposed method is faster and more effective.

Key words: small sample problem, boundary detection, texton, supervised learning, classifier

中图分类号: