基于注意力机制的旋转不变指纹细节点特征提取网络  

Rotation Invariant Fingerprint Detail Point Feature Extraction Network Based on Attention Mechanism

作　　者：苏毅婧 Su Yi-jing(Quanzhou Institute of Equipment Manufacturing,Haixi Institutes,Chinese Academy of Sciences,Quanzhou 362216,Fujian Province,China)

机构地区：[1]中国科学院海西研究院泉州装备制造研究中心,福建泉州362216

出　　处：《科学与信息化》2024年第2期85-87,共3页Technology and Information

基　　金：福建省自然科学基金资助项目,项目名称:指纹图像姿态估计算法研究,项目编号:2020J05094。

摘　　要：准确提取指纹图像的细节点特征是实现指纹精确匹配的关键步骤。细节点特征通常以任意方向出现,而普通的卷积神经网络并没有明确地对旋转变换进行建模。为解决上述问题,本文提出一种基于注意力机制的旋转不变指纹细节点特征提取网络。该模型在模型设计上加入对旋转的权重共享,从而降低模型的参数量,从根本上增强了模型的泛化能力,提高了计算效率。本研究为构建稳定高效的指纹细节点检测系统提供了新思路。Accurately extracting the details of fingerprint images is a key step to achieve accurate fingerprint matching.Detail point features often appear in any direction,and ordinary convolution neural networks do not explicitly model rotational transformations.In order to solve the above problems,this paper proposes a rotational invariant fingerprint detail point feature extraction network based on attention mechanism.In this model,the weight sharing of rotation is added to the model design,so as to reduce the number of model parameters,fundamentally enhance the generalization ability of the model,and improve the computational efficiency.This study provides a new idea for constructing a stable and efficient fingerprint detail point detection system.

关 键 词：指纹特征提取 深度学习 注意力机制 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]