摘要
提出了一种基于多尺度LBP特征的人脸描述与识别算法。对原始人脸图像进行二级小波分解,并采用LBP算子分别计算两幅低频逼近图像的LBP特征谱;将LBP特征谱划分为若干个互不重叠的特征区域,然后分别进行直方图统计。最后,将所有区域的LBP直方图序列连接起来得到多尺度LBP特征,将其作为人脸的鉴别特征用于分类识别。所提出算法在ORL人脸数据库中取得了高达99%的人脸识别率。实验分析表明,多尺度LBP特征具有较强的人脸图像描述能力和可鉴别性,且对人脸表情及位置的变化具有较高的鲁棒性。
In order to improve the accuracy and robustness of face recognition, a face description and recognition method based on multi-scale Local Binary Pattern（LBP） feature is proposed. The original face image is decomposed into two levels by wavelet analysis, and the LBP operator is applied to two approximate images respectively to extract LBP feature map. Then, the two maps are divided into several regions,in which the histograms are computed and linked to get a multi-scale LBP feature. Fi- nally, the multi-scale LBP feature is used as the face descriptor for classification and recognition. The experimental results on ORL face database show that the proposed method can achieve high face recognition rate up to 99% ,which shows that the multi-scale LBP feature has highly descriptive and discriminable abilities for human face and is robust to face expressions and position variations.
出处
《光学精密工程》
EI
CAS
CSCD
北大核心
2008年第4期696-705,共10页
Optics and Precision Engineering
基金
重庆市自然科学基金资助项目(No.CSTCNo.2006BB2152)
关键词
人脸识别
多尺度分析
LBP算子
直方图
face recognition
multi-scale analysis
Local Binary Pattern（LBP） operator
histogram
作者简介
王玮（1979-），男，福建晋江人，博士，主要从事模式识别、图像处理、嵌入式系统等方面的研究。E-mail：weipub@163．com
