Gender and Age Categorization Using Gaze Analysis

This paper is concerned with effective and efficient identification of human subjects for the purpose of data mining and personalized service, e.g., retail space characterized by clutter. We expand on the recently introduced Gaze ANalysis Technique for human identification (GANT) to show its potential use for gender and age (younger or older than 30 years) categorization. The motivation for gaze analysis comes from its direct relation to some target of interest, with behavioral biometrics including the length of time for observing the object of interest and the scanning pattern involved. The scan paths are represented by a graph structure (with nodes and weighted arcs) that encodes duration of a fixation (how long the observer fixated the same point), densities (how close points fixated by the observer are), and arcs (the graph adjacency matrix). Two different categorization protocols to detect demographics from human gaze are used, average distance and Adaboost. Data is captured using the Tobii 1750 remote eye tracker, which integrates all its components (camera, infrared lighting, etc.) into a 17’’ LCD monitor (1280×1024 resolution). An image sensor then records pupil position and corneal reflections to determine eyes position and the gaze point. The biometric dataset captured consists of 72 males and 39 females. Age ranges between 17 and 80 years, with 43 subjects “under 30” and 68 subjects “over 30”. The target set subjects gaze at consists of close-up faces of 8 males and 8 females, with half of the faces (4 males and 4 females) belonging to celebrities (mostly actors and actresses), while the others are of people unknown to the testers.