Gaze-Based Graphical Password Using Webcam

Abstract

Authentication refers to verification of the identity of an user. There exist various types of authentication techniques, starting from simple password based authentication up to behavior biometric based authentication. In this paper, a new way of authentication is proposed where the user provides her password through eye gaze. It is based on graphical password scheme where she can choose her password from a large image data set. At the time of authentication, she needs to recall it and look at the chosen passcode appearing in a display in correct sequence. The method uses a machine learning technique where a convolutional neural network is used to determine gaze locations using inputs from a simple web camera. It takes her cropped eye as input and provides gaze location as output. Proposed method is cost effective solution as gaze tracking is done through a simple web camera. The proposed method is also free from attacks such as shoulder surfing, smudge, brute-force attacks. Experiments have been carried out to validate the system. It has been observed to perform accurately for all the volunteers.