使用掌纹线对基于深度学习的掌纹识别进行数据增强  

Using palmprint lines for data enhancement of palmprint recognition based on deep learning

作　　者：金怡凡 王海涛 贾伟[1] JIN Yifan;WANG Haitao;JIA Wei(School of Computer Science and Information Engineering,Hefei University of Technology,Hefei 230009,China)

机构地区：[1]合肥工业大学计算机与信息学院,安徽合肥230009

出　　处：《智能系统学报》2024年第5期1178-1189,共12页CAAI Transactions on Intelligent Systems

基　　金：国家自然科学基金项目(62076086).

摘　　要：近年来掌纹识别技术受到越来越多的关注,然而在掌纹识别的过程中,复杂的应用场景为识别带来了困难。在基于深度学习的掌纹识别算法中,数据增强操作具有较大的作用。由于掌纹的独特性,其所包含的特征信息几乎全部处于掌纹线之中,因此传统的全局数据增强方法收效甚微。本文提出了一种基于掌纹线的数据增强方法。该方法首先基于传统的图像处理方法,提出了多阶段的掌纹线提取算法。然后,基于提取的掌纹线,设计了一种掌纹识别的数据增强方案。通过实验表明,应用该数据增强方式对掌纹图像进行增强之后,在4个广泛应用的深度学习模型上都取得了更好的效果。该数据增强方法简单高效,能够在实际应用中发挥作用。In recent years,palmprint recognition technology has attracted growing attention.However,complex application scenarios bring difficulties in the process of palmprint recognition.Data enhancement plays an important role in palmprint recognition algorithms based on deep learning.Owing to the uniqueness of palmprint,nearly all the feature data it contains lie within the palmprint lines.As a result,traditional global data enhancement methods have little effect in this case.In this study,a data enhancement method based on palmprint lines is proposed.In the method,a multi-stage palmprint line extraction algorithm is first proposed on the basis of the traditional image processing method.Subsequently,a data enhancement scheme on palmprint recognition is designed on the basis of the extracted palmprint lines.Experiments demonstrate that applying this data enhancement method to improve palmprint images has achieved better results than four widely used deep-learning models.The data enhancement method is simple and efficient and can play a role in actual applications.

关 键 词：生物特征识别 深度学习 身份鉴别 GABOR滤波器 掌纹识别 掌线提取 数据增强 卷积神经网络 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]