基于孪生网络的掌纹识别方法
Palmprint Recognition Method Based on Siamese Networks
摘要
本文基于孪生网络提出了一种掌纹识别方法，先通过手部图像数据进行简单的二分类任务对网络卷积层进行预训练以提升网络特征提取能力，再更换网络全连接层进行自监督学习以获得输入图像的低维特征向量，最终通过特征向量间的欧氏距离确定掌纹所有者身份，无需使用额外数据进行微调即可实现精确的掌纹身份识别。经实验证明，该方法具有较高的准确率与稳定性，在自建数据集上可达到98%，在掌纹识别任务上优于多种于常见的图像分类网络。Abstract
This paper proposes a palmprint recognition method based on Siamese networks. Firstly, hand image data is used to perform a simple binary classification task to pretrain the network\'s convolutional layers and enhance the network\'s feature extraction capability. Then, the fully connected layers of the network are replaced for self-supervised learning to obtain low-dimensional feature vectors of input images. Finally, the identity of the palmprint owner is determined by calculating the Euclidean distance between feature vectors, eliminating the need for fine-tuning with additional data while achieving accurate palmprint identification. Experimental results demonstrate that the proposed method achieves high accuracy and stability, with a recognition rate of up to 98% on a self-built dataset, outperforming several commonly used image classification networks in palmprint recognition tasks. 关键词
模式识别/掌纹识别/孪生网络/图像分类Key words
Pattern recognition/Palmprint recognition/Siamese networks/Image classification引用本文复制引用
蒲秋梅,黄波,田景龙.基于孪生网络的掌纹识别方法[EB/OL].(2023-09-22)[2025-12-14].http://www.paper.edu.cn/releasepaper/content/202309-50.学科分类
计算技术、计算机技术
评论