A component-based approach to hand-based verification and identification system.

摘要

Hand-based verification/identification represent a key biometric technology with a wide range of potential applications both in industry and government. Traditionally, hand-based verification and identification systems exploit information from the whole hand for authentication or recognition purposes. To account for hand and finger motion, guidance pegs are used to fix the position and orientation of the hand. In this dissertation, we have investigated a component-based approach to hand-based verification and identification which improves both accuracy and robustness as well as ease of use due to avoiding pegs. Our approach accounts for hand and finger motion by decomposing the hand silhouette in different regions corresponding to the back of the palm and the fingers. To improve accuracy and robustness, verification/recognition is performed by fusing information from different parts of the hand. The proposed approach operates on 2D images acquired by placing the hand on a flat lighting table and does not require using guidance pegs or extracting any landmark points on the hand. To decompose the silhouette of the hand in different regions, we have devised a robust methodology based on an iterative morphological filtering scheme. To capture the geometry of the back of the palm and the fingers, we employ region descriptors based on high-order Zernike moments which are computed using an efficient methodology. The proposed approach has been evaluated both for verification and recognition purposes on a database of 101 subjects with 10 images per subject, illustrating high accuracy and robustness. Comparisons with related approaches involving the use of the whole hand or different parts of the hand illustrate the superiority of the proposed approach. Qualitative and quantitative comparisons with state-of-the-art approaches indicate that the proposed approach has comparable or better accuracy. As an extension of our work, we investigate the problem of gender classification from hand shape. It has been motivated by studies in anthropometry and psychology suggesting that it is possible to distinguish between male and female hands by considering certain geometric features. For classification, we compute the distance of a given part from two different eigenspaces, one corresponding to the male class and the other corresponding to female class. We have experimented using each part of the hand separately as well as fusing information from different parts of the hand. Using a small database containing 20 males and 20 females, we report classification results close to 98% using score-level fusion and LDA. Also, we address the template aging issue. We introduce a technique by decomposing the hand silhouette into the different parts and analyzing the confidences of these parts in order to lead to global optimization of templates. In the proposed method, first the hand silhouette is divided in different parts corresponding to the fingers. Then the confidence of each finger, as well as its identity, is evaluated by a Support Vector Data Description (SVDD). The confidence of a query hand is determined by the maximum confidence of all fingers. If the maximum confidence is higher than a threshold, the boundaries of all fingers' SVDDs are incrementally updated to learn the variations of the input data. The motivation behind this technique is that the temporal changes that may occur in the fingers are uncorrelated in such a way that the confidence of each finger can be significantly different from the others. As a result those fingers with difficult intra-class variations can be used in the update process by this technique. The experimental results show the effectiveness of the proposed technique in comparison to the state of the art self-update technique specially at low false acceptance rates.