Typing Biometrics: Impact of Human Learning on Performance Quality
Typing Biometrics: Impact of Human Learning on Performance Quality
复制标题
打字生物识别技术：人类学习对表现质量的影响
DOI：
--
复制
发表时间：
2011
期刊：
影响因子：
--
通讯作者：
M. Tremaine
中科院分区：
文献类型：
--
作者：
Benjamin K. Ngugi;Beverly K. Kahn;M. Tremaine
使用被盗的个人身份信息（如社会安全号码）进行身份欺诈仍然是一个主要问题。冒名顶替者可以通过拥有被盗的标识信息来冒充真正的用户，这是当前认证系统的一个弱点。向传统知识和基于令牌的身份验证系统添加生物识别层是解决这个问题的一种方法。物理生物识别技术，如指纹系统，是高度准确的;因此，它们将是此类应用的首选，但往往是不合适的。行为生物识别技术，如生物识别类型模式，有可能填补这一空白，作为另一个层次的安全性，但这项研究发现了一些性能质量的缺陷。出现了两个改进的研究流。第一种方法试图通过构建更好的分类器来提高性能，而第二种方法试图通过使用更丰富的识别输入来实现相同的目标。两个流都假设打字生物特征模式随时间推移是稳定的。本研究通过分析学生打字模式随时间的变化来探讨这一假设的有效性。结果表明，打字模式随着时间的推移而变化，由于学习导致几个性能质量的挑战。首先，变化的模式导致认证精度下降。其次，在训练期间创建的参考生物特征模板的相关性变得有问题。第三，准确性的恶化损害了整个系统的安全性，第四，净效应带来了生物特征键盘是否不再“适合使用”作为认证系统的问题。如果行为生物识别技术要在最大限度地减少身份验证欺诈方面发挥重要作用，这些都是需要解决的关键数据质量问题。可能的解决方案的问题，包括生物模板更新和选择不相关的PIN组合，建议作为未来研究的潜在课题。
The use of stolen personal-identifying information, like Social Security numbers, to commit identity fraud continues to be a major problem. The fact that an impostor can pass as the genuine user by possession of stolen identification information is a weakness in current authentication systems. Adding a biometric layer to the traditional knowledge and token-based authentication systems is one way to counter this problem. Physical biometrics, such as fingerprint systems, are highly accurate; hence, they would be the first choice for such applications but are often inappropriate. Behavioral biometrics, like biometric typing patterns, have the potential to fill this gap as another level of security but this research identified some deficiencies in performance quality. Two research streams for improvements have emerged. The first approach attempts to improve performance by building better classifiers, while the second attempts to attain the same goal by using richer identifying inputs. Both streams assume that the typing biometric patterns are stable over time. This study investigates the validity of this assumption by analyzing how students’ typing patterns behave over time. The results demonstrate that typing patterns change over time due to learning resulting in several performance quality challenges. First, the changing patterns lead to deteriorating authentication accuracy. Second, the relevancy of the reference biometric template created during training becomes questionable. Third, the deterioration in accuracy compromises the security of the whole system and fourth, the net effect brings to question whether the biometric keypad is no longer “fit for use” as an authentication system. These are critical data quality issues that need to be addressed if behavioral biometrics are to play a significant role in minimizing authentication fraud. Possible solutions to the problem, including biometric template updating and choice of uncorrelated PIN combinations, are suggested as potential topics for future research.