Use of intelligent system techniques for storage and retrieval of biometrics data with application to personal identification

Abstract

Biometrics (or biometric recognition) refers to the technology recognizing individuals based on their physiological and/or behavioral characteristics. It has advantages over token-based and knowledge-based personal recognition technologies. The important biometric systems including face, fingerprint, iris and palmprint are in fact image-based recognition systems. For identification purpose, these systems can be regarded as Content-Based Image Retrieval (CBIR) systems. The storage and retrieval requirements of image-based biometric systems, nevertheless, make them to be a special type of CBIR systems. First, personal physiological characteristics cannot be properly described by traditional text-based methods. Second, there is no generally accepted indexing and classification methods being used in common biometrics systems. Third, the captured biometric images always contain noise and variations due to change in capture environments and user habit. Fourth, in real world applications, they should be scalable for large databases and should provide relevant security mechanisms. Finally, they may make its own decision or provide help to make decision on whether a claim is accepted or rejected. In this research, we focus on the storage and retrieval of some image-based biometric systems supported by personal physiological characteristics. We have considered the storage and retrieval of biometric templates as an application of target and category search in narrow domain. We have incorporated some primitive visual-based image features including texture, lines and points, as content descriptors of biometric images. Various intelligent system techniques, e.g. machine learning and fuzzy, have been used to extract and represent the visual-based image features of palmprint and face images. We have developed methods that can compactly represent and effectively retrieve palmprint and face templates and we are the first to consider retrieval in large palmprint databases. We originally identify and study three design issues of cancelable biometrics, which is a security and privacy enhancement method proposed for template protection. We are among the first to consider the three issues integrally when designing and evaluating cancelable biometrics.