Abstract: Since the user’s typing pattern has been evaluated as a stable and unique characteristic to some extent, keystroke dynamics is a well known technique in security research. The traditional keystroke authentication methods usually use keystroke features such as the digraph latency extracted from two consecutive keystrokes and the keystroke duration. However, these traditional algorithms do not take the changing rate of any two adjacent keystroke features into consideration, which may degrade the performance of keystroke dynamics identity classification. In order to solve the above problems, a new concept of diversity degree of keystroke feature curve is firstly proposed in this paper and applied to the identity classification. The novel keystroke dynamic identity authentication algorithm not only uses the traditional keystroke features, but also introduces the changing rate of any two adjacent keystroke features, so it can recognize different user’s keystroke behavior much better than traditional methods. The experimental results demonstrate the effectiveness of the proposed algorithm, compared with several algorithms including Manhattan, Manhattan（scaled）, statistics, neural networks and machine learning.

Key words: biometric authentication, identity classification, keystroke feature, fixed text, keystroke feature curve

摘要： 击键特征是一种能反映用户行为的动态特征，可作为用户身份识别的信息源。传统的认证方法通常仅采用击键特征向量中所包含的每个特征值的大小来进行身份识别，而没有利用任意两个相邻特征值之间的变化率，在一些情况下，可能导致识别准确度不理想。针对上述问题，定义了一种新颖的击键特征曲线差异度的概念，并由此提出基于击键特征曲线差异度的认证算法。该认证算法不仅利用了常规的击键特征信息，还首次引入了任意两个相邻特征值之间的变化率信息，使算法性能得到显著提高。实验结果表明，相比于曼哈顿距离算法、统计学算法、神经网络算法和机器学习算法，新算法的错误拒绝率、错误接受率和相等错误率更低，识别准确度更高，效果更好。

关键词: 生物认证, 身份识别, 击键特征, 固定文本, 击键特征曲线