基于人脸特征和AdaBoost算法的多姿态人脸检测.pdfVIP

第 30卷第 4 期 计算机应用 Vol. 30 No. 4 2010年 4 月 　 Journal of Compu ter App lication s 　 Ap r. 20 10 文章编号 : 100 1 - 9081 (2010) 04 - 0967 - 04 基于人脸特征和 A daBoo st算法的多姿态人脸检测 阮锦新 ,尹俊勋 (华南理工大学 电子与信息学院 ,广州 51064 1) ( ruanj inxin@ 163. com ) 摘 　要 :基于人脸特征和 A daBoo st算法 ,提出一种改进的多姿态人脸检测算法 。首先利用肤色特征快速排除绝 大部分背景区域 ,然后在肤色区域中搜索眼睛和嘴巴区域 ,根据眼睛和嘴巴区域的几何特征所确定的人脸方向分割 出大致正向的人脸候选区域 ,最后利用 A daBoo st算法对候选区域进行分类 。实验表明 ,算法能实现多姿态人脸的快 速检测 ,而且对脸部表情和遮挡有较强的鲁棒性 。 关键词 :人脸检测 ;多姿态 ;肤色区域 ;人脸特征 ; A daBoo st算法 ; H arrlike特征 中图分类号 : TP39 1. 4 1　　文献标志码 : A M ultipose face detection ba sed on fac ia l fea tures and A daBoost a lgor ithm RUAN J inxin , Y IN Junxun (S chool of E lectron ic and Inf orm a tion Eng ineering, S ou th Ch ina Un iversity of Technology, Guangzhou Guang dong 51064 1, Ch ina) A b stract: A n imp roved m u ltipo se face detection algorithm based on facial features and A daBoo st algorithm wa s p ropo sed. M ak ing fu ll u se of facial sk in color inform ation firstly, mo st of the background region s cou ld be qu ick ly excluded. A fter detecting eye s and mou th, the app roxim ate fron tal face candidate region s were segm ented according to face orien tation decided by the geom etric feature s of the eyes and mouth region s. A t last, the face candidate region s were cla ssified by A daBoo st algorithm. The exp erim ental re su lts demon strate that the algorithm can further imp rove the m u ltipo se face detection accu racy and is h igh ly robu st to facial exp ression and occlu sion. Key words: face detection; m u ltipo