Multi-Biometric System


Some high-level biometric systems combine, or fuse, several biometrics to increase performance over that of the individual biometric systems. The "Continual Burst Authentication Strategy" section of [1] gives an overview of what we are trying to do.

Project Description

This project will combine several biometric systems - two that we have developed and one under development. Initially, the keystroke and stylometry systems will be combined and performance measured, and some preliminary work in this area has been initiated by John Stewart [2]. Later, we will also combine the mouse movement system being developed in project 2.

Convert keystroke sequences into text

We are currently capturing data through a keylogger that captures keyboard and mouse data. However, it does not have access to sent email or Word documents, so if we want stylometry information (linguistic level word and syntax information) we must obtain it from the keystroke data. Therefore, the text data for stylometry analysis will be obtained by converting keystroke input back into text, and it is possible to come close to getting it correct except for mouse editing. For example, the input of characters and spaces presents no problem. Strikes of the "backspace" and "delete" keys, however, will delete data and and these data must be appropriately deleted. You must keep track of where the cursor is currently located because it can be moved, for example, by the "home", "end", and arrow keys prior to hits of keys that delete data. This will require a relatively simple program. The input is a file of keystroke data and the output is a file of text.

Browser Data Experiments

Browsing on the Internet, such as performing Google searches, usually involves both keyboard and mouse input. Therefore, browser data appears to be ideal for combining keystroke, mouse, and possibly stylometry information. Browser data will be provided by Ned Bakelman. Team 1 will provide the mouse feature vectors and your conversion program will provide the text to obtain the stylometry feature vectors. Ned and your customers can help you run the experiments. Also, Vinnie is working on a new version of the biometric classifier and might run experiments for you. The following experiments are of interest:

Key References

  1. John V. Monaco, Ned Bakelman, Sung-Hyuk Cha, and Charles C. Tappert, Developing a Keystroke Biometric System for Continual Authentication of Computer Users, EISIC Conf., Denmark, 2012.
  2. J.C. Stewart, J.V. Monaco, S. Cha, and C.C. Tappert (2011). An Investigation of Keystroke and Stylometry Traits. Proc. Int. Joint Conf. Biometrics (IJCB 2011), Wash. D.C., October 2011.

Pace University Biometric Authentication System

This is a general authentication system that operates on feature vectors where each vector component is a real number in the range 0-1. In contrast to identification (1-of-n problem), authentication is a binary problem (yes, you are the person you claim to be, or no you are not).