基于神经网络的手指三模态多级特征编码融合方法

Fusion Method of Finger Trimodal Multilevel Coded Features Based on NN

摘要: 本文提出了一种基于神经网络的手指三模态多级特征编码融合方法。首先，针对手指三模态原始图像分别构建轻量级卷积神经网络实现多级特征提取。其次，利用聚合向量编码方式对手指三模态多级特征进行编码，分别得到指静脉编码特征、指纹编码特征和指节纹编码特征。然后，利用构建的全连接神经网络模型分别将手指三模态浅层特征和深层特征进行融合。最后，将手指三模态浅层融合特征和深层融合特征进行串联融合，得到手指三模态融合特征。实验结果表明通过该方法得到的融合特征的识别准确率为99.76%，说明该方法能够得到具有良好个性特征表达能力的融合特征，可以有效提高个体身份识别精度。

Abstract: In this paper, a neural network based fusion method of finger trimodal multi-level coded features is proposed. Firstly, lightweight convolutional neural networks are constructed respectively for original images of finger trimodal to extract multi-level features. Secondly, coding the finger trimodal multi-level features based on local aggregate vector theory, finger-vein coding features, fingerprint coding features and finger-knuckle coding features are obtained respectively. Then, the finger trimodal shallow coded features and deep coded features are respectively fused using a fully connected neural network. Finally, the finger trimodal shallow fusion features and deep fusion features are fused in series. The experimental results show that the recognition accuracy of fusion features obtained by this method is 99.76%, which indicates that this method can obtain fusion features with good expression ability of personality features, and can effectively improve the accuracy of individual identity recognition.