基于SVM算法的实时人脸验证的研究  

Research on real-time face verification based on SVM algorithm

作　　者：叶文武 成杰 高颂[1] 徐玮巍 张强[1] 徐寅林[1] Ye Wenwu;Cheng Jie;Gao Song;Xu Weiwei;Zhang Qiang;Xu Yinlin(School of Physics and Technology,Nanjing Normal University,Nanjing 210023,China)

机构地区：[1]南京师范大学物理科学与技术学院,江苏南京210023

出　　处：《国外电子测量技术》2018年第12期85-90,共6页Foreign Electronic Measurement Technology

摘　　要：论文在OpenCV的开发环境中设计了基于SVM算法的实时人脸验证系统,能够进行是否是同一人的分类,分析了SVM模型误差来源,优化了模型参数,运用Dlib工具和归一化方法,克服了姿态的变化时模型的鲁棒性较差,且在细节特征较少时识别的准确率不高的问题。最终在以自建的504组相同人脸和不同人脸为正负样本训练集上训练出的SVM分类器,在300组的相同人脸和不同人脸的测试样本样本集取得了89.7%的准确率。It is designed a real-time face verification system based on SVM algorithm in OpenCV development environment. It can classify whether it is the same person, analyze the error source of SVM model, optimize the model parameters, use Dlib tool and normalization method to overcome. When the posture changes, the robustness of the model is poor, and the accuracy of recognition is not high when the detailed features are less. Finally, the SVM classifier trained on self-built 504 groups of identical faces and different faces as the positive and negative sample training sets achieved 89.7% accuracy of the same set of face and different faces of 300 test samples.

关 键 词：支持向量机 人脸验证 OPENCV 

分 类 号：TN919.8[电子电信—通信与信息系统]