摘要
由于高分辨率掌纹图像尺寸大、掌纹图像数量少等特点,目前主流的方法采用细节特征点匹配,其算法设计复杂,识别精度不高。针对以上问题,本文提出基于迁移学习的高分辨率掌纹图像识别方法,该方法以VGG16为基础网络,将在imagenet数据集上训练好的权重参数,用于初始化所有的卷积层;使用图像增强技术将高分辨率掌纹图像分别4、9、16、25等分,采用投票的方法得到整个掌纹图像的准确率,最高可达到99.69%。经实验证明,该方法可以实现端到端的高精度高分辨掌纹图像识别,识别率优于以往的基于细节特征点匹配方法。
Due to the large size of high-resolution palmprint images and the small number of palmprint images,the current mainstream method adopts detail feature point matching,which has complicated algorithm design and low recognition accuracy. To solve the above problems,a high resolution palmprint image recognition method based on transfer learning is proposed in this paper.This method takes VGG16 as the base network and initializes all convolutional layers with weight parameters trained on ImageNet data set. The high resolution palmprint images are divided into4, 9, 16 and25 equal parts by image enhancement technology,and the accuracy of the whole palmprint image is obtained by voting method,which can reach up to99.69%. Experimental results show that this method can realize end-to-end high precision and high resolution palmprint image recognition,and the recognition rate is better than the previous method based on detail feature point matching.
作者
吴碧巧
邢永鑫
王天一
WU Biqiao;XING Yongxin;WANG Tianyi(College of Big Data and Information Engineering,GuiZhou University,Guiyang 550025,China)
出处
《智能计算机与应用》
2021年第5期37-42,共6页
Intelligent Computer and Applications
基金
贵州省科技厅与贵州大学科技合作计划项目(黔科合LH字[2016]7431号)。
作者简介
吴碧巧(1994-),女,硕士研究生,主要研究方向:深度学习、图像识别;邢永鑫(1993-),男,硕士研究生,主要研究方向:深度学习、目标检测;通讯作者:王天一(1989-),男,博士,副教授,主要研究方向:量子通信、图像处理、计算机视觉,Email:tywang@gzu.edu.cn。
