Assessment of geometric features for individual identification and verification in biometric hand systems

Abstract

This paper studies the reliability of geometric features for the identification of users based on hand biometrics. Our methodology is based on genetic algorithms and mutual information. The aim is to provide a system for user identification rather than a classification. Additionally, a robust hand segmentation method to extract the hand silhouette and a set of geometric features in hard and complex environments is described. This paper focuses on studying how important and discriminating the hand geometric features are, and if they are suitable in developing a robust and reliable biometric identification. Several public databases have been used to test our method. As a result, the number of required features have been drastically reduced from datasets with more than 400 features. In fact, good classification rates with about 50 features on average are achieved, with a 100% accuracy using the GA–LDA strategy for the GPDS database and 97% for the CASIA and IITD databases, approximately. For these last contact-less databases, reasonable EER rates are also obtained.

Introduction

User identification based on Biometrics is a very mature technique used in many industrial applications including airports and IT systems. Biometrics are based on the measurement of characteristics which are unique to an individual, invariable through time, not invasive and difficult to counterfeit. Such characteristics include: iris, fingerprint, face, voice and hands (Jain, Ross, & Prabhakar, 2004). In this paper we focus on hand based Biometrics.

Despite the maturity of biometrics based identification systems, there is still a large amount of research taking place. Human hands can be used to build biometric systems based on fingerprints, palm prints, vein patterns and hand geometry. Hand geometry identification systems are getting a considerable importance in medium security applications because they take advantage of several factors which other biometrics traits do not (Sanchez-Reillo, Sanchez-Avila, & Gonzalez-Marcos, 2000). From the hardware viewpoint, just low resolution cameras are needed, and the computational cost of the applied algorithms is relatively scarce, so it is easily accepted by the users. Hence, hand-based biometric systems have attracted and motivated the interest of a large number of researchers for the last years (Duta, 2009, Fong and Seng, 2009). Previous studies concentrate on the use of hand based biometric systems for user classification. This paper concentrates on the use of these systems for user identification.

In this approach, geometric features are used since they are simpler and easier to compute. Unlike other papers in the literature which try to extract many features as possible and apply subsequently matching fusion techniques (Nanni and Lumini, 2009, Ross and Jain, 2003), our objective is to reduce as much as possible the number of features while maintaining the same rates of individual identification and recognition. Thus, the selected features are easier to interpret, which implies we are able to analyse which of them are more important as discriminant biometrical features. Additionally, it could be further extended to determine which hand regions are really suitable to provide good descriptors.

Genetic algorithms (GAs) are also considered as suitable evolutionary strategies for feature selection in the literature. They are well adapted for problems with a large number of features (Raymer, Punch, Goodman, Kuhn, & Jain, 2000), and are applied to different areas, from object detection (Sun, Bebis, & Miller, 2004) to gene detection in microarray data (McLachlan, Bean, & Peel, 2002). With this kind of feature selection approach it is expected to achieve a threefold objective: improving the accuracy rate of the classifiers; providing faster and more cost-effective predictors because of a significant reduction of the number of features; and providing a better understanding of the underlying process that generated the data. Furthermore, we combine computational intelligence techniques with statistical approaches such as mutual information (Guyon & Elisseeff, 2003), with the aim of providing feedback about whether the subsets generated are equivalent in terms of the correlation among the variables of each subset.

The remainder of this paper is organised as follows: Section 2 provides a detailed description of existing methods for user identification based on hand geometry. Section 3 presents the image processing techniques used to extract from the hand images whereas Section 4 describes the steps needed to achieve robust and reliable geometric hand features. Section 5 sets out the methodology of this approach and Section 6 shows the experimental results over several well-known public databases. A discussion about the usefulness and significance of the selected features over all the databases is found in Section 7. Section 8 concludes the article.