计算机工程与应用 ›› 2018, Vol. 54 ›› Issue (7): 132-137.DOI: 10.3778/j.issn.1002-8331.1610-0308

• 模式识别与人工智能 • 上一篇    下一篇

抗外点干扰的鲁棒AdaBoost分类器构建方法

曹万鹏,罗云彬,史  辉   

  1. 北京工业大学 未来网络创新中心,北京  100124
  • 出版日期:2018-04-01 发布日期:2018-04-16

Robust AdaBoost classifier construction method against outlier interference

CAO Wanpeng, LUO Yunbin, SHI Hui   

  1. Advanced Innovation Center for Future Internet Technology, Beijing University of Technology, Beijing 100124, China
  • Online:2018-04-01 Published:2018-04-16

摘要: 根据AdaBoost算法易受外点影响这一缺陷,提出一种利用Ransac算法实现抗外点干扰的鲁棒AdaBoost分类器构建方法。不同于其他AdaBoost算法在分类器构建中单纯使用样本加权或权值控制的手段,该算法将Ransac算法引入AdaBoost分类器模型构建过程中,去除潜在外点,克服现有AdaBoost算法缺陷。同时,借助Ransac算法,从全部AdaBoost分类器中选择最佳分类器模型,消除由外点引起的分类器降级。最后,将该AdaBoost分类器模型用于含有一定量外点的笔迹样本进行验证,实验结果证明了该方法的有效性。

关键词: AdaBoost分类器, Ransac算法, 样本加权, 分类

Abstract: Taking this reason that AdaBoost is sensitive to outliers, robust AdaBoost classifier is constructed against outlier interference using Ransac. Different from the other weak classifier sample weighting and controlling method in AdaBoost, Ransac is employed and introduced to the process of classifier model construction to overcome the drawbacks of the existing AdaBoost weak classifier weighting algorithms. Meanwhile, the adverse affection of outliers can be effectively eliminated by virtue of the Ransac algorithm’s strong ability in removing outliers. Through above strategy, the classifier degradation is able to be avoided. Finally, in the validation experiment, the designed classifier model is applied in the handwriting samples classification including some outliers. The experimental results show its validity.

Key words: AdaBoost classifier, Ransac algorithm, sample weighting, classification