基于深度学习的小面积指纹匹配方法

发布时间：2018-05-17 13:09

本文选题：指纹匹配 + 深度学习　；参考：《计算机应用》2017年11期

【摘要】：针对传统的基于细节特征点的指纹匹配方法多适用于采集面积较大的指纹,在面向智能手机端的小采集面积指纹时准确率明显下降的问题,提出一种基于深度学习的小面积指纹匹配方法。首先,提取指纹图像的细节特征点信息;其次,搜索和标定感兴趣纹理区域(ROI);然后,构建并改进基于残差结构的轻量级深度神经网络,通过采用二值化特征模式优化网络和Triplet Loss方式训练模型;最后,制定一种智能手机端注册匹配策略实现小面积指纹匹配。实验结果表明,提出方法在公开库FVCDB1与自建数据库上的等错率(EER)分别仅为0.50%与0.58%,远低于传统的基于细节特征点的指纹匹配方法,能够有效提升小面积指纹匹配的性能,更好地满足智能手机端的应用需求。
[Abstract]:The traditional fingerprint matching method based on detail feature points is suitable for fingerprint with large area, and the accuracy of fingerprint with small area for smart phone is obviously decreased. A small area fingerprint matching method based on depth learning is proposed. Firstly, the minutiae of fingerprint image are extracted; secondly, the texture region of interest is searched and calibrated; then, the lightweight depth neural network based on residual structure is constructed and improved. The binary feature pattern is used to optimize the network and Triplet Loss training model. Finally, a smart phone registration matching strategy is developed to achieve small area fingerprint matching. The experimental results show that the proposed method is only 0.50% and 0.58% in the open library FVCDB1 and the self-built database respectively, which is far lower than the traditional fingerprint matching method based on detail feature points, and can effectively improve the performance of small area fingerprint matching. Better meet the application needs of smart phone.
【作者单位】：浙江工业大学计算机科学与技术学院;浙江外国语学院计算机科学与技术学院;杭州易和网络有限公司;
【基金】：浙江省自然科学基金资助项目(Y1101304)~~
【分类号】：TP18;TP391.41

【相似文献】