Pré-processamento dos intervalos de tempos observados na dinâmica de digitação (KeyStroke) de senhas curtas

Abstract

In 2006, a method was proposed concerning the use of time interval equalization to improve performances of some biometric methods based on typing dynamics (or keystroke). In the paper where that method was first proposed, relatively small databases were used for showing, in terms of error rates, the effect of time equalization applied as a preprocessing step before the use of classical methods. In 2009, a much larger large database for keystroke research was made publicly available, through the Internet. This database is based on a single hypothetical password, typed by 51 volunteers through 8 sessions (50 samples per session). In this dissertation, the preprocessing method is adapted to this large public database of short typing patterns. Thus, by using the same biometric detectors already used by the owners of the database, we obtain new experimental results which clearly show an outstanding performance gain when the equalization interval (preprocessing) is applied. It is also studied the performance gain as a function of the password length (in number of symbols), and the stability of typing pattern against changes in the order of typed symbol pairs. Finally, the last study was carried to a new database we acquired with both direct and inverted sequence of symbols, allows for the analysis of keyboard layout changes on biometric performances.