一种结合AdaBoost与mean shift的人眼检测方法.doc

文档介绍：
一种结合 AdaBoost 与 mean shift 的人眼检
测方法
熊金水1,苏菲1,张建2**
5
10
(1. 北京邮电大学信息与通信工程学院,北京 100876;
2. 北京邮电大学宣传部,北京 100876)
摘要:人眼检测是人脸识别系统的中间模块,对人脸识别系统的整体性能影响很大。在系统
中,人眼检测被分为粗定位和后处理两个部分。粗定位部分使用 Haar 特征、AdaBoost 算法
和级联结构,获得大量的候选点。后处理部分使用含尺度因子的 mean shift 算法,找到概率
密度估计最大的位置,并作为最终精确的人眼位置。实验表明,对于正面人脸,该算法能实
时精准的找到人眼位置。
关键词:人脸识别;人眼检测;AdaBoost;Mean shift
中图分类号:
15
Eye detection based bination of AdaBoost and mean
shift
XIONG Jinshui1, SU Fei1, ZHANG Jian2
(1. School of Information munication Engineering, Beijing University of Posts and
20
25
30
35
40
munications, Beijing 100876;
2. Propaganda Department of Beijing University of Posts and munications, Beijing
100876)
Abstract: Eye detection is an important part of the face recognition system. It plays a pivotal role
in the accuracy of face recognition. In the system, eye detection is divided into coarse detection
and fine detection. The haar feature, the AdaBoost algorithm and the attentional cascade are used
in the coarse detection to get plenty of candidates. The mean shift, including scale factor, are used
in the fine detection to find the maximum position of probability density, which is considered as
the ultimate eye position. The experiments show that the algorithm can localize precise eyes for
frontal faces in real time.
Key words: Face recognition; Eye detection; AdaBoost; Mean shift
0 引言
人脸识别系