基于深度学习的人脸识别系统

Face Recognition System Based on Deep Learning

为了提高人脸识别的精度和性能,基于深度学习算法设计并实现了一种实时的人脸识别系统,并分析该系统完成人脸识别任务的基本流程。构建的人脸识别系统采用MTCNN(多任务级联神经网络)作为人脸检测算法,融合KNN(K最邻近分类算法)的FaceNet人脸识别方法,利用FaceNet进行人脸表征,基于KNN进行人脸特征分类。对设计的人脸识别系统分别进行识别率测试、响应时间测试、复杂环境干扰测试;结果表明:系统准确率达到98%,在线平均响应时间为0.67 s,在眼、嘴、耳、鼻部位遮挡比例为50%的环境下识别成功率平均约为72%,验证了系统的可行性及算法的有效性。

In order to improve the accuracy and performance of face recognition,this paper proposes to design and implement a real-time face recognition system based on deep learning algorithm,and analyze the process of completing facial recognition tasks.The proposed face recognition system uses MTCNN(Multi-task Cascaded Neural Network)as the face detection algorithm,and integrates the FaceNet face recognition method of KNN(K-Nearest Neighbor Classification Algorithm).FaceNet is used for facial representation,and facial features are classified based on KNN.Recognition rate testing,response time testing,and complex environmental interference testing are conducted on the proposed face recognition system.The results show that the recognition accuracy of the system reaches 98%,with an online average response time of 0.67 seconds,and the recognition success rate reaches 72%on average in an environment where the occlusion ratio of the eyes,mouth,ears,and nose is 50%.The feasibility of the system and the effectiveness of the algorithm are thus verified.