基于分数层融合的多生物特征融合识别  

Multi-biometric feature fusion recognition based on fractional layer fusion

作　　者：李小敏 陈英[1] LI Xiaoming;CHEN Ying(School of Software,Nanchang Hangkong University,Nanchang 330063,China)

机构地区：[1]南昌航空大学软件学院,江西南昌330063

出　　处：《长江信息通信》2021年第10期7-11,共5页Changjiang Information & Communications

基　　金：国家自然科学基金(61762067)。

摘　　要：针对多生物信息识别分类问题,提出了一种基于分数层融合的掌纹和虹膜融合识别模型。首先使用1D LogGabor滤波及最小汉明距离匹配实现了虹膜的特征提取和识别匹配,识别准确度达到98.9%;其次利用优化后的SqueezeNet网络模型实现了掌纹的分类识别,其分类准确效率可达99%;最后采用分数层融合方案按掌纹与虹膜比为4:6的权重比进行多生物融合识别,最终实现识别分类准确度为99.75。此外,设定评价指标对掌纹、虹膜以及融合后的识别性能进行了评估,得到该三个识别系统的AUC值分别为0.994875、0.985471、0.999599。实验结果表明,多模态生物特征融合识别有效地提高了系统识别的性能,使其具有更高的识别效率和准确度,在安全性、可靠性和鲁棒性等方面都有所增强。Aiming at the problem of multi-biological information recognition and classification,a palmprint and iris fusion recognition model based on fractional layer fusion is proposed.Firstly,1D log Gabor filtering and minimum Hamming distance matching are used to realize iris feature extraction and recognition matching,and the recognition accuracy reaches 98.9%;Secondly,the optimized Squeezenet network model is used to realize palmprint classification and recognition,and the classification accuracy and efficiency can reach 99%;Finally,the multi organism fusion recognition is carried out according to the weight ratio of palmprint to iris of 4:6,and the final recognition accuracy is 99.75%.In addition,the performance of palmprint recognition,iris recognition and fused recognition are evaluated by setting evaluation indexes,and the AUC values of the three recognition systems are 0.994875,0.985471 and 0.999599 respectively.The experimental results show that the multimodal biometric fusion recognition can effectively improve the performance of the system recognition,make it have higher recognition efficiency and accuracy,and enhance the safety,reliability and robustness.

关 键 词：虹膜识别 掌纹识别 融合识别 多模态生物特征 分数层融合 

分 类 号：G391.41[文化科学]