CAJ | 학술논문

鉴于人脸与人耳两种生物特征在图像获取上的相似性以及生理位置上的互补性，提出一种人脸-人耳多模态融合模板保护方法，将两者在特征层进行融合，然后利用模糊保险箱算法对融合模板进行保护。该方案基本流程分为五部分：图像预处理、Gabor-PCA特征提取、特征融合、融合模板加密以及融合模板解密。在由ORL人脸库和USTB人耳库3构成的人脸-人耳多模态图像库上的认证实验结果表明所提模板防护方法的有效性，且基于融合模板保护的认证结果比基于单模板保护的认证结果在识别率和误识率上均有所优化。
감우인검여인이량충생물특정재도상획취상적상사성이급생리위치상적호보성，제출일충인검-인이다모태융합모판보호방법，장량자재특정층진행융합，연후이용모호보험상산법대융합모판진행보호。해방안기본류정분위오부분：도상예처리、Gabor-PCA특정제취、특정융합、융합모판가밀이급융합모판해밀。재유ORL인검고화USTB인이고3구성적인검-인이다모태도상고상적인증실험결과표명소제모판방호방법적유효성，차기우융합모판보호적인증결과비기우단모판보호적인증결과재식별솔화오식솔상균유소우화。
Due to biological characteristic similarity in image acquisition and physiological complementarity of the face and the ear, a face and ear multimodal fusion template protection method was proposed, in which the face and the ear were combined in the feature level and then a fuzzy vault was utilized to protect the fusion feature template. The basic flow of the scheme was divided into 5 parts: image preprocessing, Gabor- PCA feature extraction, feature fusion, fusion template encryption, and fusion template decryp-tion. On the multimodal image dataset consisted of the ORL face dataset and the USTB ear dataset 3, authentication experimental re-sults show the effectiveness of the proposed fusion template protection method. Also the fusion template protection method outperforms the unimodal template protection method on both genuine accept rate and false accept rate.