基于级联支持向量机融合多特征的人脸检测  

FACE DETECTION BASED ON CASCADE SUPPORT VECTOR MACHINE FUSING MULTI-FEATURE

作　　者：张小龙[1] 刘书炘[2] 刘满华[1] 

机构地区：[1]上海交通大学仪器科学与工程系,上海200240 [2]漳州师范学院教育科学与技术系,福建漳州363000

出　　处：《计算机应用与软件》2016年第4期151-154,207,共5页Computer Applications and Software

摘　　要：人脸图像中包含丰富的特征信息,不同特征具有其各自的优势。基于此,提出一种基于级联支持向量机有效融合多种特征的人脸检测算法。该算法首先利用肤色模型对待检图像进行预处理,筛选出疑似人脸区域。然后在疑似区域中提取图像的HOG(Histogram of Oriented Gradients)和LBP(Local Binary Patterns)特征,并分别对这两种特征集进行特征选择,训练两个SVM(Support Vector Machine)分类器,最后将两个SVM分类器级联起来实现人脸检测。在多个人脸图像数据库上的实验结果表明,该人脸检测算法提高了人脸检测率,降低了误检率,并且对多种光照条件、姿态、表情以及部分遮挡的情况都具有较好的鲁棒性。Face images contain rich features information,and different features have their own advantages. This paper proposes a face detection algorithm,which is based on effective fusion of multiple features with cascade support vector machines. The algorithm first uses complexion model to preprocess the detecting image and screens out suspected human face region. Then,it extracts HOG（ histogram of oriented gradients） and LBP（ local binary patterns） features of image from the suspected face region,and makes feature selection on these two feature sets respectively,and trains two SVM（ support vector machine） classifiers as well. Finally,the algorithm cascades these two SVM classifiers to implement face detection. Results of experiment on multiple face databases show that the proposed algorithm improves the face detection rate and reduces the false detection rate,and has high robustness for the conditions of a variety of illumination,posture,facial expressions,and partial occlusion.

关 键 词：人脸检测 HOG特征 LBP特征 SVM分类器 级联分类器 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]