Chen Chi-Jim, Pai Tun-Wen, Cheng Mox
Department of Computer Science and Engineering, National Taiwan Ocean University, Pei-Ning Road, Keelung 20224, Taiwan.
Egis Technology Inc., 2F, No. 360, Rueiguang Road, Neihu District, Taipei 11492, Taiwan.
Sensors (Basel). 2015 Mar 31;15(4):7807-22. doi: 10.3390/s150407807.
A sweeping fingerprint sensor converts fingerprints on a row by row basis through image reconstruction techniques. However, a built fingerprint image might appear to be truncated and distorted when the finger was swept across a fingerprint sensor at a non-linear speed. If the truncated fingerprint images were enrolled as reference targets and collected by any automated fingerprint identification system (AFIS), successful prediction rates for fingerprint matching applications would be decreased significantly. In this paper, a novel and effective methodology with low time computational complexity was developed for detecting truncated fingerprints in a real time manner. Several filtering rules were implemented to validate existences of truncated fingerprints. In addition, a machine learning method of supported vector machine (SVM), based on the principle of structural risk minimization, was applied to reject pseudo truncated fingerprints containing similar characteristics of truncated ones. The experimental result has shown that an accuracy rate of 90.7% was achieved by successfully identifying truncated fingerprint images from testing images before AFIS enrollment procedures. The proposed effective and efficient methodology can be extensively applied to all existing fingerprint matching systems as a preliminary quality control prior to construction of fingerprint templates.
一种扫描式指纹传感器通过图像重建技术逐行转换指纹。然而，当手指以非线性速度扫过指纹传感器时，采集到的指纹图像可能会出现截断和失真的情况。如果将截断的指纹图像作为参考目标录入并由任何自动指纹识别系统（AFIS）采集，指纹匹配应用的成功预测率将显著降低。本文提出了一种新颖且有效的方法，该方法具有较低的时间计算复杂度，能够实时检测截断的指纹。实施了若干过滤规则来验证截断指纹的存在。此外，基于结构风险最小化原则的支持向量机（SVM）机器学习方法被用于排除具有类似截断特征的伪截断指纹。实验结果表明，在AFIS录入程序之前，从测试图像中成功识别截断指纹图像的准确率达到了90.7%。所提出的有效且高效的方法可广泛应用于所有现有的指纹匹配系统，作为构建指纹模板之前的初步质量控制手段。
