基于质量评价与特征提取网络的手指静脉识别

Finger Vein Recognition Based on Quality Evaluation and Feature Extraction Network

在利用手指静脉信息进行个人身份识别认证的系统中,采集到的手指静脉图像质量的好坏以及图像特征提取算法的性能直接影响着系统识别的准确性。针对以上两个核心问题,提出了基于神经网络学习的手指静脉图像质量判断与特征提取技术,并将图像质量判断、特征提取与匹配集成为一个完整的识别系统。在采集手指静脉图像时,利用经过训练的轻量级的MobileNet-V2网络,判断图像质量,仅保留质量较好的图像,将其用于后续个人身份识别与验证;对特征提取网络的设计,提出了基于平滑平均准确度损失函数的ResNet模型,提高了原网络的特征提取能力。最后,利用余弦相似度进行特征匹配,获得待识别人的身份信息。实验结果表明,所提出的方法在公开的山东大学和中国香港理工大学手指静脉数据库上分别得到3.855%和3.699%的等误率值,比基于差分图像+VGG网络识别等方法至少降低了1.311%。并且,所设计的识别系统对于自建手指静脉图像数据库,达到了99.30%的识别率。

In a system that uses finger vein information for personal identification and authentication,the quality of the finger vein image and the performance of the image feature extraction algorithm directly affect the accuracy of the system recognition.In order to solve these two problems,a technique of finger vein image quality judgment and feature extraction based on neural network learning was proposed,and the image quality judgment,feature extraction and matching were integrated into a complete recognition system.When collecting finger vein images,a lightweight MobileNet-V2 network was trained to determine the image quality,and only the high-quality images were retained for subsequent personal identification and verification.For the design of feature extraction network,a Res Net model based on smooth average precision loss function was proposed,which improves the feature extraction capability of o-riginal network.Finally,cosine similarity was used for feature matching to obtain the identity information of the person to be identified.The experimental results show that the equal error rates of the proposed method are 3.855%and 3.699%respectively in the finger vein database of Shandong University and the finger vein database of Hong Kong Polytechnic University.Compared with the methods of difference image+VGG and the other three methods,the results are reduced by at least 1.311%.Moreover,the recognition rate of the designed system is 99.30%for the self-built finger vein image database.