【摘要】 为提高出入境自助通关系统中THFaceID人脸认证算法的性能,分析通关系统中的运行环境、旅客通关习惯等影响因素,针对模糊、多姿态和复杂光照下的现场人脸图像,提出一种多形状局部二值模式(multi shape local binary pattern,MS-LBP)特征,能够较好描述特殊结构的非矩形纹理,对多姿态人脸进行有效检测。使用异方差方式进行鉴别特征的提取,将现场照与备案照的特征向量变换到同一特征空间进行认证。实验结果表明,升级后核心算法的认证性能在FAR=0.001处比原版本系统有近20%的提高。更多还原

【Abstract】 To improve the performance of the THFaceID verification algorithm of automated passenger clearance system,the runtime environment and customs of passenger in the clearance system and other influencing factors were analyzed,and a multi shape local binary pattern(MS-LBP)feature was proposed to describe the special structure of non-rectangular textures.Multipose face under complex environment was effectively detected.A heterogeneous feature projection algorithm was provided to improve the feature discriminability of probe and gallery images in the verification system.Results of experiments show that the upgraded system improves the performance by 20%compared with that of the original version at FAR=0.001,which demonstrates the effectiveness of the algorithm.更多还原