计算机工程 ›› 2012, Vol. 38 ›› Issue (12): 182-184.doi: 10.3969/j.issn.1000-3428.2012.12.054

• 人工智能及识别技术 • 上一篇    下一篇

基于半监督聚类的人脸检测方法

王 燕,蒋正午   

  1. (兰州理工大学计算机与通信学院,兰州 730050)
  • 收稿日期:2011-07-22 出版日期:2012-06-20 发布日期:2012-06-20
  • 作者简介:王 燕(1971-),女,副教授,主研方向:模式识别,数据库技术,图像处理;蒋正午,硕士研究生
  • 基金项目:
    甘肃省自然科学基金资助项目(1014RJZA009)

Face Detection Method Based on Semi-supervised Clustering

WANG Yan, JIANG Zheng-wu   

  1. (School of Computer and Communication, Lanzhou University of Technology, Lanzhou 730050, China)
  • Received:2011-07-22 Online:2012-06-20 Published:2012-06-20

摘要: 将肤色与连续AdaBoost算法相结合进行人脸检测,并引入半监督策略指导肤色聚类从而建立肤色模型。在肤色聚类过程中,提出一种基于半监督的SKDK算法引导肤色聚类,依据各个像素簇的概率统计分布特性得到肤色模型。在此基础上利用数学形态学等知识对图像进行处理,得到人脸候选区域,将其作为连续AdaBoost分类器的输入进行人脸检测。实验结果表明,在多人脸的场景下,该方法的检测效果优于直接使用连续AdaBoost方法进行人脸检测的检测效果。

关键词: 人脸检测, 半监督策略, 聚类, 肤色模型, 数学形态学, 连续AdaBoost

Abstract: The paper proposes a method of face detection combined color of skin with continuous AdaBoost algorithm. In order to establish skin color model, this paper takes advantage of semi-supervised strategy to guide skin color clustering, and it also proposes a new algorithm SKDK in the process of clustering. skin color model can be established by the probability statistics distribution characteristics of each pixel cluster. On this basis, mathematical morphology of knowledge is used to handle image and find face candidate, which is the input of continuous AdaBoost classifier for final face detection. Experimental results prove that face detection ability of the method is superior to that directly using continuous AdaBoost method for face detection especially in multi-face situation.

Key words: face detection, semi-supervised strategy, clustering, skin color model, mathematical morphology, continuous AdaBoost

中图分类号: