中文摘要：

使用LBP算子对视频中的人脸图像提取特征,通过线性SVM分类器进行人种分类,并利用级联投票机制提高人种识别的正确率,大大降低了视频序列中人脸误检带来的干扰.本文提出的识别方法在FERET数据库中具有较好的性能;在人种识别自行建立的LFW和WEB复杂训练数据库中通过交叉验证测试的识别率达到91.10%;该方法在视频数据库中的平均识别率可达86.29%,大量实验证明本文方法对自然场景中的光照、角度和位置变化都具有较高的鲁棒性.

英文摘要：

In this paper,LBP (local binary pattern)descriptor was utilized to extract face image fea-ture from every video by using linear SVM (support vector machine)to classify human race.Simulta-neously,a cascading voting scheme was proposed to better improve the performance of race recogni-tion task.It greatly reduces the interference caused by the proposed method,and has a prominent high recognition rate on FERET (face recognition technology)dataset.The proposed scheme can reach 91.10% on the self-established LFW (labeled faces in the wild)and WEB dataset for race recog-nition through cross-training method and achieve 86.29% on the video dataset.The sufficient experi-ments show that the algorithm are robust to illumination variation,angle and location change.