基于三种类型图像数据的人脸识别测试.pdfVIP

刑事技术 2016 年 第41 卷 第6 期 ·论  著· DOI ：10.16467/j.1008-3650.2016.06.003 基于三种类型图像数据的人脸识别测试 1 2 1,* 1 1 3 谢兰迟 ，王俊娟 ，黎智辉 ，许 磊 ，张 宁 ，商怀哲 （1. 公安部物证鉴定中心，北京 100038 ；2. 蚌埠市公安局，安徽 蚌埠 233040 ； 3. 德州市公安局德城分局，山东 德州 253000） 摘　要：人脸识别的准确率受很多因素的影响，如图像尺寸、姿态角度、年龄变化等。本文提出了一种基于三 种类型图像数据的人脸识别测试流程。测试中，采集了 300 个测试对象的静态照片、三维人像以及监控视频资 料。采集的数据经过预处理，并按照人像的分辨率和拍摄角度分类。我们通过对 7 800 张人脸样本的测试，分 析不同分辨率、角度、年龄对于人脸识别算法鲁棒性的影响。本文提出的方法是一种更加适合法庭科学领域应 用的人脸识别系统评估与测试方法。 关键词：人像检验；人脸识别测试；监控视频；三维人像 中图分类号 ： DF793.2 文献标识码 ：A 文章编号 ：1008-3650(2016)06-0442-04 Face Recognition Test Based on Three Types of Image Data 1 2 1,* 1 1 3 XIE Lanchi , WANG Junjuan , LI Zhihui , XU Lei , ZHANG Ning , SHANG Huaizhe 1. Institute of Forensic Science, Ministry of Public Security, Beijing 100038, China; 2. Bengbu Public Security （ Bureau, Anhui, Bengbu 233040, China; 3. Decheng Sub-bureau of Public Security, Shandong, Dezhou 253000, China) ABSTRACT: The accuracy of face recognition is influenced with a variety of factors, such as the image size, subject’s pose and age. In this paper, an effective face recognition test method was proposed based on three different image sources. In the test, 300 individuals were collected of their still, 3D face images and video frames from surveillance. The captured face images were preprocessed and classified by resolution and the angle that the camera was shot to subject. Through measurement and analysis of the obtained 7800 face images, how the resolution, pose and age influence the algorithm robustness of face recognition has been explored. The method presented here is suitable for