Increasing level of confidence of iris biometric matching

A key requirement of biometric matching is to identify people with high level of confidence. This confidence level indicates a degree that the biometric system can reliably decide whether a query biometric template belongs to a registered biometric template or not. This paper explores the use of a template transformation and the longest common substring expression to tackle variability during biometric acquisition and increase the level of confidence of iris biometric matching. The transformation provided with a derivative of the registered template will initially convert the query template to a transformed template that can be used to perform an exact match. When the transformation has been performed, the longest common substring between the registered template and the transformed template is obtained. The proposed transformation will cause the intra-class distributions more homogeneous and push the distributions to a large similarity while the proposed longest common substring will push the inter-class distributions to a large dissimilarity thus minimizing the chance of false acceptances and increasing the separation between the intra-class and inter-class distributions. We extensively tested our proposed method using iris images from the commercial Bath dataset and found that the decidability index (d') and Fisher-ratio can be increased to 82.5 and 3401 respectively. Moreover, the success rate of the transformation to produce exact matches is 96.4%.