A hamming distance-like filtering in keystroke dynamics

Keystroke dynamics-based user authentication is recognized as one of potentially valid biometrics. For this research, we asked 51 subjects to input predefined text and collected their keystroke data. For two data, their dissimilarity is measured, with which they are verified to be collected from the same subject or not. In this paper, we show that dissimilarity measures have their EER depend on filtering of outlier for collected data. To verify that two data are collected from different subjects, we propose a new filter method based on the resemblance to Hamming distance. Authentication experiments say that our proposal is valid.