Investigation of infrared hand vein pattern biometrics

The increasing public awareness of security issues has led to a massive rise in demand for high accuracy and reliable personal identification systems. Many biometric systems have been developed to meet the requirements. Many human surface features such as the faces and fingerprints have been studied extensively to identify a person for a number of years. Whilst these features can offer relatively high accuracy, their tolerance to forgery is moderate. There is a recent developing trend of studying the unique interior features (e.g. iris, retina etc.) of the human body for biometric purposes. Amongst those interior features, human vein patterns offer the potential for robust performance, and, have not been fully investigated previously by researchers. This thesis investigates the potential of the hand vein patterns serving as a biometric feature. Several major research issues are identified and investigated regarding the hand vein patterns as biometrics. These research issues are: (i). Collectibility of the vein pattern images; (ii). Processing of the vein pattern images; (iii). Identification of features for matching the vein patterns; (iv). The study of the discriminating power of vein patterns in biometric applications. In order to carry out the study, a model is proposed with five independent stages for the vein pattern biometric system, namely, Data Acquisition, Preprocessing, Vein Pattern Segmentation, Skeletonization and Matching. Various imaging technologies are reviewed before the Infrared imaging technique is selected to construct a hand vein pattern image database. Several novel image processing algorithms are proposed to enhance the image quality and extract the vein pattern skeleton from the raw images. A novel feature, the Line Edge Maps, is proposed to analyze the vein patterns, and achieved encouraging results in the application of personal verification. Further study was carried out on the similarity of vein pattern between identical twins. With limited data, the experiment found the hand vein patterns between identical twins can be genetically independent; and they are distinguishable both visually and computationally. Overall, the study provides evidence demonstrating the vein pattern image data is obtainable in a non-intrusive manner; and some unique computational features can be extracted from the vein pattern images after appropriate pre-processing, and these features are capable of providing strong discriminating power in the context of biometric applications.DOCTOR OF PHILOSOPHY (SCE