基于局部特征提取的形变指纹深度学习检测网络  

Altered Fingerprint Deep Learning Detection Based on Local Feature Extraction

作　　者：徐超 冯国瑞[1] 

机构地区：[1]上海大学通信与信息工程学院,上海200072

出　　处：《工业控制计算机》2022年第10期103-105,108,共4页Industrial Control Computer

基　　金：国家自然科学基金(62072295)。

摘　　要：形变指纹能通过改变指纹的纹理信息躲避指纹识别系统的检测。基于深度学习的形变指纹检测方法,考虑指纹本身的结构特征,提出了两个可以嵌入网络任何部位的局部特征提取模块,将其添加到深度学习模型中:一个是局部关键特征,代表了指纹局部区域的显著特征点;另一个是局部关联特征,代表了局部区域内各个特征点之间的关联程度。提取到的两个局部特征与全局特征融合后,继续在网络中训练,提高形变指纹的检测结果。将提出的模块添加到多个网络中进行测试,验证了模块的有效性。Altered fingerprint change its texture to avoid the detection of fingerprint identification system. This paper studies the altered fingerprint detection based on deep learning method. Taking the structural features of the fingerprint,two local feature extraction modules which can embedded in any part of the network are proposed and added to the deep learning method. One is the local key feature, which represents the most important feature points in the local area of the fingerprint. The other is the local correlation feature, which represents the correlation between the feature points in the local area. The two local features will be fused with the global feature and continue to be trained in the network to improve the detection of altered fingerprints. The module is added to multiple networks for comparison, which proves the effectiveness of the module.

关 键 词：形变指纹检测 局部关键特征 局部关联特征 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP18[自动化与计算机技术—计算机科学与技术]