作　　者： ; ; ;

机构地区： 北京大学深圳研究生院

出　　处： 《华中科技大学学报（自然科学版）》 2013年第S1期116-120,共5页

摘　　要： 基于Haar特征的Adaboost人脸检测算法,提出一种简单的人脸多窗口合并方法.在保证较高精度的检测率的情况下对检测出的人脸进行预处理,采用空间信息与纹理特征相结合的方法,即人脸图像分块与改进的局部二元模式相结合的方法进行人脸特征提取.用支持向量机(SVM)作为分类器进行性别分类,并实时在视频中标记,实现了视频序列中实时人脸检测和性别识别.实验结果表明:在室内室外多种复杂场景的非高清视频序列中,检测人脸的速度可以达到20帧/s,人脸性别识别部分可以在每秒内识别30个以上人脸,满足视频序列当中实时性检测和高准确率识别的要求. A real time face gender recognition framework that is capable of processing images rapidly in video sequences was descrobed.The Adaboost algorithm was used based on Haar features for face detection of high precision and a new simple way was presented to combine detected face windows. The method of combining the spatial information and the texture features which is LBP(local binary pattern)was used in feature extraction for gender recognition.Finally,the extracted features were used to train SVM(support vector machine)classifiers.Experimental results in an ordinary video sequences in a variety of indoor and outdoor complex scenes show that face detection speed can be achieved at 20frames per second and gender recognition result at 30faces per second.Thus,it satisfies the demand of high detection rate and real time recognition in the video stream.

关 键 词： 图像识别 人脸检测 性别识别 算法 局部二元模式 支持向量机

领　　域： [自动化与计算机技术] [自动化与计算机技术]