基于总错误率和特征关联的自适应融合多模态生物特征识别  

Multibiometric recognition based on total error rate and feature correlation adaptive fusion

作　　者：吴迪[1] 唐勇奇[1,2] 林国汉[1] 胡慧[1] 

机构地区：[1]湖南工程学院电气信息学院,湖南湘潭411004 [2]湖南风电装备与电能变换协同创新中心,湖南湘潭411004

出　　处：《光电子．激光》2016年第1期67-76,共10页Journal of Optoelectronics·Laser

基　　金：国家科技支撑计划(1214ZGA008);国家自然科学基金(61263031;61563030);湖南省自然科学基金(14jj6041);湖南省教育厅(14k029;15A044);重庆市教委自然科学基金(KJ1400628);湖南工程学院博士科研启动基金(15045)资助项目

摘　　要：针对单生物特征识别准确率和鲁棒性差的问题,提出了一种基于总错误率(TER)和特征关联自适应融合多模态生物特征识别方法。首先将TER作为判别特征引入到多模态识别,以代替传统的匹配分数;其次在不确定度量理论的基础上,考虑人脸特征和语音特征之间的时空关联性,提出了一种基于特征关联的多特征自适应融合策略,利用特征关联系数自适应调节不同识别特征对识别结果的贡献。仿真实验表明,与几种代表性的融合算法相比,本文所提出的融合模式可以有效提高多生物特征识别系统的准确性和鲁棒性。On account of the bad accuracy and robustness of unimodal biometric recognition,a novel approach is proposed based on total error rate（TER）and feature correlation adaptive fusion method.This study aims at voiding calculating posterior probability of a certain score transfers scores into TERbased on corresponding false reject rates（FRRs）and false accept rates（FARs）at first.Secondly,a new adaptive fusion method named based on information entropy theory and uncertainty measurement theory is proposed,which considers the correlation between face information and speech information,and the new adaptive fusion method adjusts the fusion weights adaptively between weight sum fusion scheme and multiplicative fusion scheme using the characteristic of correlation coefficient.Experimental results show that the proposed method outperforms the traditional score level fusion schemes,the experimental result also confirms the effectiveness of the proposed method to improve the performance of multibiometric system is confirmed.

关 键 词：总错误率(TER) 错误拒绝率(FRR) 错误接收率(FAR) 特征关联系数 自适应融合 多生物特征识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]