一种基于多尺度LBPH特征的快速人脸识别算法  

THE FAST FACE RECOGNITION ALGORITHM BASED ON MULTI-SCALE LBPH FEATURE

作　　者：朱峰[1] 

机构地区：[1]上海城市管理职业技术学院人文与信息技术学院,上海200438

出　　处：《计算机应用与软件》2015年第8期315-319,共5页Computer Applications and Software

摘　　要：人脸探测和相似度比较的相关算法经过十几年的发展,目前已经进入实用的阶段。在目前先进研究成果的基础上,侧重考虑处理速度和识别性能的均衡性,实现实时的人脸识别系统。首先基于Haar特征快速提取人脸轮廓,然后基于形变组件模型训练、提取脸部关键部位的landmark,并根据landmark采用多尺度LBPH(Local Binary Patterns Histogram)表示人脸特征,采用距离尺度学习算法训练尺度模型并进行人脸识别。系统人脸识别时间接近实时,识别性能接近实用水平。After the development over ten more years,the correlated algorithms of face detection and similarity comparison have been stepped into the stage of practical application. Based on current advanced research results,we focus on considering the balance between processing speed and recognition performance,and realise the real-time face recognition system. The system first fast extracts the face contour based on Haar features. Then based on deformable part model it trains and extracts landmarks of facial key part. According to landmark it uses multi-scale LBPH（ local binary patterns histogram） to represent facial features,and uses distance metric learning algorithm to train the scale model and to recognise faces. The face recognition time of the system closes to the real-time,and the recognition performance closes to the practical level as well.

关 键 词：人脸探测 脸部landmark提取 距离尺度学习 人脸识别 形变组件模型 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]