基于仿生模式识别的多镜头人脸身份确认系统研究  

Multi-Camera Human-Face Personal Identification System Based on the BIOMIMETIC PATTERN RECOGNITION

作　　者：王守觉[1] 徐健[1] 王宪保[2] 覃鸿[1] 

机构地区：[1]中国科学院半导体研究所神经网络实验室,北京100083 [2]浙江工业大学智能信息系统研究所,杭州310014

出　　处：《电子学报》2003年第1期1-3,共3页Acta Electronica Sinica

基　　金：国家自然科学基金 (No 60 1 350 1 0 )

摘　　要：提出了一种利用仿生模式识别原理以多镜头信息融合的人脸身份确认方法 .讨论了以多镜头与多次采样建立仿生模式识别多权值神经元网络的基本理论 .并介绍了以三个镜头五次采样作训练样本的实验系统 ,以及人脸预处理过程步骤 .实验结果表明在确保无误认的前提下 ,正确确认率达 96 % ,漏认率 (即拒认率 )为 4 % .实验中并对比显示了增加神经元网络复杂度提高识别效果的作用 .A Multi-camera Human-face Personal Identification method, based on the Biomimetic Pattern Recognition, had been proposed. The basic theory of neural networks with Multi-weight Neuron, in Multi-camera Human-face Personal Identification System Based on the Biomimetic Pattern Recognition, had been discussed. A three-camera system takes five combined pictures as learning samples in experiments. The basic course of human-face picture preprocessing had been described. The results of experiments showed the correct identification rate as 96%, rejection rate as 4%, and no mistake in identification. Experiments showed that increasing the complexity of neural networks makes better identification result.

关 键 词：仿生模式识别 人脸识别 身份确认 数据融合 神经网络 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]