计算机工程

• 人工智能及识别技术 • 上一篇    下一篇

基于Adaboost与Clifford代数的人脸检测

杨晋吉,李荣兵   

  1. (华南师范大学计算机学院,广州 510631)
  • 收稿日期:2012-03-19 出版日期:2013-09-15 发布日期:2013-09-13
  • 作者简介:杨晋吉(1968-),男,教授、博士,主研方向:自动推理,模式识别;李荣兵,硕士研究生
  • 基金项目:
    广东省自然科学基金资助项目(10151063101000031)

Face Detection Based on Adaboost and Clifford Algebra

YANG Jin-ji, LI Rong-bing   

  1. (School of Computer, South China Normal University, Guangzhou 510631, China)
  • Received:2012-03-19 Online:2013-09-15 Published:2013-09-13

摘要: Adaboost算法在光照不均、背景复杂的条件下进行人脸检测时误检率较高。为解决该问题,提出一种基于Adaboost算法与Clifford代数矢量积性质的人脸检测方法。利用Adaboost算法初步定位人脸可能存在的区域,对该区域进行基于知识的校验,如果校验失败,根据Clifford矢量积性质,寻找与待验证区域相似度较高的人脸,当相似度大于阈值时,判断其为人脸。实验结果表明,与Viola-Jones方法相比,该方法在保持较高检测率的同时,降低了误检率,且鲁棒性较好。

关键词: 人脸检测, Adaboost算法, Clifford代数, 矢量积, 主成分分析, 人脸先验知识

Abstract: In the conditions of complicated backgrounds and different illumination, as face detection based on Adaboost algorithm usually has higher false alarm rate, a new method based on the Adaboost algorithm and the Clifford vector product is proposed in this paper. Most of the non-face region is quickly excluded by the Adaboost classifier. The candidate region is verified basing on the face prior knowledge. If verification failure, according to Clifford vector product properties, searching for the region which has higher similarity with the region that need to be verified again, when their vector product is higher than threshold, this paper can judge that it is a face region. The comparison of this method with Viola-Jones method, experimental result shows that this method can detect face with high detection rate, suppresses the error detection rate, and is highly robust to face detection.

Key words: face detection, Adaboost algorithm, Clifford algebra, vector product, Principal Component Analysis(PCA

中图分类号: