CNN Algorithm For Gaze-Based Password Identification
Abstract: Based on a previously-developed eye-tracking system which captures eye movements, and records location of eye center in real-time, a touch-free password authentication approach had been described. This touch-free password authentication system is based on detected eye center locations as the eye gazes at a variety of characters on a simple keypad over time. The preliminary method of password identification involved matching eye center points on a 3×3 keypad grid and identifying those grids in which eye center points cluster together. This method suffers from inaccuracies due to eye closure, or wandering eyes during capture, which could result in erroneous character detection and password identification. This method is also limited when the keypad contains more than nine characters as in the 3×3 keypad, and, therefore, the grid for each character is smaller.
In this work, a more robust gaze-based password identification method is developed to classify the characters gazed by the user. A Convolutional Neural Network (CNN) algorithm has been developed to perform classification of the gazed characters. The method utilizes the multidimensional information available from the eye center data that includes not only the horizontal and vertical coordinates of the gazed location, but also the temporal information captured during eye tracking. Temporal information is inherent in the time stamp of the analyzed frames during real-time eye tracking. The CNN algorithm is developed that allows the system to learn with the data and progressively improve its performance. The CNN model is applied to the human gaze sequences captured from a 4×4 keypad which offers a larger pool of characters for a password. In this presentation, we will share the results for hands-free password identification for a 5-character password from a 4×4 keypad using CNN algorithm.
Judging Forms – Official judges only