Biometrics, which refers to automatic recognition of persons based on their distinctive physiological and behavioral characteristics, is an application of the fields of image processing and pattern recognition. Most of the readers of this newsletter are familiar with biometrics, as technical articles on the topic have appeared regularly in IAPR and IEEE publications over the past three decades. It is somewhat surprising that the first monograph on the subject, The Guide to Biometrics, would publish as late as in 2004, but it is a pleasure to find that it is well written. This monograph has a very good consistency of terminology and notation and avoids repetition of content among various chapters – qualities that are hard to find in edited volumes. Another great strength of this book is that it discusses common fundamentals underlying the entire biometrics field instead of concentrating on individual biometric characteristics or implementations. The authors of this book are distinguished researchers in the field of biometrics; their outstanding background and firm grasp of the subject is evident throughout the book. Along with a discussion of the underlying concepts, the authors have shared numerous practical insights and suggestions from their experience throughout the book, which is very valuable for practitioners in the field.
The book is divided into four parts. Part I is introductory; subsequent parts get increasingly sophisticated. Depending on their expertise, readers can choose to selectively read any part. The book contains an extensive bibliography and an index, which makes it valuable as a reference text. In Chapters 1 and 2, the authors acquaint the reader with the common terminology used in biometrics and their meanings, e.g., the authors explain the difference between verification, identification, screening, and continuity of identity systems. The authors illustrate how pattern recognition centric probabilistic biometric matching fundamentally differs from computer security centric exact password matching. While password-based verification systems have several limitations (common passwords can be easily guessed, cryptic passwords can be easily forgotten, and all passwords can be shared), their information content (i.e., cryptographic strength) can be arbitrarily increased (at the expense of inconvenience to the user in remembering a long cryptic string). On the other hand, the cryptographic strength (accuracy) offered by biometric systems is limited by the inherent information content in the biometric characteristic, and the accuracy of the feature extraction and classification methods used by the system. In spite of three decades of research, designing highly accurate biometric systems remains a very challenging pattern recognition research problem. In Chapters 3 and 4, while introducing the basic concepts underlying common (fingerprint, face, speaker, iris, hand geometry, and signature) and emerging (DNA, retina, thermograms, gait, keystroke, ear, skin reflectance, lip motion, and body odor) biometric systems, the authors guide the readers through the difficulties and challenges one would encounter in designing an accurate system for any of these biometric characteristics. This compels the reader to read the rest of the book.
In Part II, the basic concepts on biometrics performance estimation introduced in Part I are expanded and formalized with mathematical notation. False match rate, false non-match rate, false accept rate, false reject rate, receiver operating characteristic curve, equal error rate, d-prime, failure to acquire, and failure to enroll, are first explained for biometric verification in Chapter 5. These are then extended to biometric identification performance matrices of reliability, selectivity, recall, precision, identification false accept rate, and identification false reject rate in Chapter 6. Chapter 6 is certain to enhance a reader’s understanding of biometric identification, which is a much harder problem than verification. Estimating real-world performance from laboratory evaluation of biometric system has proven to be elusive, especially since the error rates are very small and depend upon a large number of factors such as demographics, weather/illumination, user behavior, and application. In Chapter 7, the authors point out that the best strategy to evaluate biometric systems is to organize competitions among biometric systems and conduct comparative evaluations on a common database and testing protocol. The authors summarize the implications of biometric error rates and quote range of error rates of the state-of-the-art biometric systems from public comparative tests. No single biometric is suitable for all applications; the choice of a biometric depends upon not only the system error rates but also numerous other factors. In Chapter 8, the authors, based on their vast experience, provide guidelines to the reader on which factors to consider and how to select a suitable biometric for their application. The pros and cons of each biometric technology are discussed and some of the prevalent myths are debunked.
Part III of this book reads like a ‘practitioner’s guide’ and deals with system issues associated with implementing biometric systems. Enrollment is a significant part of a practical biometric system and the authors have devoted Chapter 9 to discuss enrollment issues such as positive and negative enrollment, enrollment for screening, quality control of enrollment data, enrollment integrity, etc. Many system level issues arise when the database size is very large. In case of certain biometric applications, such as national identification card or driver’s license, the database could contain tens of millions of identities and hundreds of millions of biometric samples. In Chapter 10, the authors walk the readers through large-scale system issues such as implication of matcher error rates, throughput, and exception handling. Among applications that require extremely accurate biometric matchers and very high throughput, there is vigorous interest in the community to use multi-modal biometric systems. Decision and score level fusions are discussed in Chapter 11 for verification and identification systems. Common design techniques used to secure end-to-end biometric systems are discussed in Chapter 12, and standards for APIs and databases, certification and legislation in Chapter 13.
Part IV of this book covers advanced topics that would be of immense interest to graduate students and experts working in the field. Estimating the inherent information content present in a biometric characteristic has profound consequences in understanding the theoretical upper bounds of achievable accuracy and, in the case of fingerprints, value as legal evidence. In Chapter 14, the authors present model-based theoretical estimation of individuality in iris and fingerprint biometrics. As mentioned earlier, biometric system evaluation is empirical. Empirical error rates may not be very accurate when the sample size is small and error rates are very small, which is very common in biometric systems. Therefore, it is important to report the statistical significance of estimates error rates. As discussed in Chapter 15, the most common method of reporting the statistical significance is through confidence intervals, which can provide indication of generalization of estimated error rates (in a probabilistic sense). Confidence interval estimation could be parametric or non-parametric. Since parametric methods are based on certain assumptions (e.g., distribution of matching scores follows Normal distributions and are independently, identically distributed), which may often be violated, authors discuss these methods only briefly. The nonparametric methods are then discussed in detail, in particular, the highly successful subset bootstrap method of confidence interval estimation. In Chapter 16, the authors discuss how cost functions can be used to choose an operating point of a matcher. Finally, methods of estimating the cumulative match curve (in identification systems) and its relation to the false accept/reject rates are discussed.
The value of this book as a textbook could have been strengthened with a summary, discussion, future research trends, and student exercises at the end of each chapter. Since Chapter 17 of the book discusses future trends, this may at best be a minor weakness. Overall, this is an excellent technical book on biometrics, capturing the common basic fundamentals underlying biometrics technology. It is poised to become a classic in this rapidly evolving field.
The Guide to Biometrics
Authors: Ruud M. Bolle, Jonathan H. Connell, Sharathchandra Pankanti, Nalini K. Ratha and Andrew W. Senior
Springer Professional Computing, 2004
Review by Salil Prabhakar