Robust Hand Geometry Measurements for Person Identification using Active Appearance Models

The increased demand for tighter border and building security has renewed public interest in biometric identification and verification systems. With fingerprint recognition being socially stigmatized, hand geometry-based recognizers have emerged as niche solutions. However, systems currently available in the marketplace require direct contact with the device, raising, among others, significant hygiene concerns. In this paper we introduce a novel approach to hand geometry-based identification. The proposed method employs Active Appearance Models to track the hand inside the capture device and to extract geometry features for identification. The AAM fitting algorithm runs faster than real-time, enabling robust system performance. In experiments on a small-scale database of hand images, the accuracy of our system exceeds 90% using as little as five features.