Segmentation of hand from cluttered backgrounds for hand geometry biometrics

Abstract

While hand geometry trait has been widely used to perform biometric recognition, majority of the methods employ images acquired against a uniform background. If segmentation of the hand is implemented, existing techniques can be used in cluttered backgrounds as well. This paper presents an approach for accurate segmentation of human hands for images following the aforementioned conditions using skin detection and shape characteristics. This technique has been developed specifically for hand geometry based authentication, and thus requires that the hand is facing the camera with the fingers spread, as required by most of the hand geometry based techniques. For skin detection, we determined HSV and RGB color ranges and further modified those values by incorporating color information from face. We used a two-step shape filtering: the first one using shape characteristics such as solidity, eccentricity, while the second one is a novel method based on the distribution of skin pixels for the hand. The algorithm also determines the location of the wrist and segments the hand above the wrist. The proposed system can be implemented in smart buildings for contactless and low-cost biometric recognition.