• Tutorials will be held on Tuesday 4th of June.

• There will be a morning session and an afternoon session

Latent Fingerprint and Palmprint Recognition

Jianjiang Feng (U. Tsinghua - China)

Fingerprint recognition is one of the important tools used in forensics to establish the identity of a suspect of criminal. Although fingerprint evidence is being widely used in forensics for more than a century, forensic fingerprint identification continues to be a semi-automated process with significant human involvement in feature extraction, matching, and decision making stages. To reduce the workload of latent fingerprint examiners and establish the error rates associated with forensic fingerprint matching, there is a critical and urgent need for fully automated forensic fingerprint identification systems. However, developing such a system is a challenging task due to: (i) poor quality of the fingerprint images acquired in forensic applications, (ii) large size of the database to be searched, and (iii) noncooperation of the criminal. This tutorial will provide an introduction to state-of-the-art tools for latent fingerprint/palmprint enhancement, latent fingerprint/palmprint matching, and quality assessment of latent fingerprints.

Sparse Representation and Biometrics

Allen Yang (UC Berkeley - USA)

The recent vibrant study of sparse representation and compressive sensing has led to numerous groundbreaking results in pattern recognition and computer vision. In this tutorial, we will present a series of three talks to provide a high-level overview about its theory, algorithms and broad applications to pattern recognition and biometrics. We will also point out ready-to-use MATLAB toolboxes available for participants to further acquire hands-on experience on these related topics.

Evidential Evaluation in Forensic Biometrics

J. González-Rodríguez & D. Ramos-Castro (U. Autónoma de Madrid - Spain)

In recent years, biometric technologies have achieved levels of performance that allow successful system deployments even in very challenging conditions. However, forensic interpretation in court from biometric samples usually rely in identifications/exclusions by human experts, as it happens in fingerprints AND most identification-of-the-source areas. Moreover, the use of biometric systems for evidence interpretation is not straightforward, since the system scores (e.g., those produced by an AFIS search) or their comparison to a threshold are not adequate to logically address the questions faced in court. In order to address logical interpretation and evaluation of the evidence using biometric systems, in this tutorial we present a Bayesian framework based on likelihood ratios. The tutorial will cover from an introduction of the nature of forensic casework and the necessary elements for a proper interpretation of the evidence, to details about how to report in court the results and validation process of a given automatic biometric system. After a review of classical forensic evidence reporting, the shortcomings and consequences of current identification procedures will be highlighted. In order to improve the scientific methods in forensics, from sample collection to evidence reporting, forensic identification disciplines must be driven by scientific procedures, resulting in transparency and testability of the techniques in use. Some statistical methods to compute likelihood ratios using biometric systems in forensic casework will also be introduced. Together with a logical framework for inference of identity, new requirements for evidence admission in Court, and the reference of DNA profiling as the new golden standard in Forensic Science to be emulated, a paradigm shift is required in order to comply with all those new requisites for XXIst century Forensic Identification Sciences. Under this likelihood ratio framework, measuring performance is critical, since there are many degrading factors affecting the computation of the evidential weight. Among them, we can cite the variability of the biometric samples among different sessions; the sparsity of the features extracted; or the degradation of the quality of the biometric samples. In these situations, likelihood ratio methods may yield misleading evidence to the court, and as a consequence could lead the fact finder to taking a wrong decision in a case. Therefore, performance is capital in order to establish the validity of the methods for its use in forensic casework, and its assessment is the subject of interest concerning the admissibility of methods in legal systems around the world. In this tutorial we introduce classical performance metrics in decision theory such as ROC or DET curves, and we show that they are insufficient to assess the performance of likelihood ratios. Therefore we introduce other metrics such as Tippett Plots, and the more recently proposed log-likelihood-ratio cost (Cllr) and Empirical Cross-Entropy (ECE). These metrics not only consider the discriminating power of the methods, but also the calibration of the likelihood ratios, which measures the adequacy of the probabilistic interpretation of the likelihood ratio in a forensic decision framework. For all the presented metrics, several implementations as software toolkits in Matlab(TM) exist, which will be available for the attendees.

Biometric Template Protection: A Unified View of Fundamental Concepts, System Architectures, and Research Challenges

Shantanu Rane (Mitsubishi Electric Research Lab)

Biometric Template Protection refers to a class of authentication or identity verification schemes that prevent users' biometric measurements from being compromised when biometric systems are attacked. This is an exciting area of research and myriad template protection proposals have appeared in the last few years. The current state of the art owes its existence to various fields: signal processing, cryptography, information theory and others. Each of these fields has provided a different language of describing the key concepts, different implementation methods, and a different way of characterizing accuracy, robustness, security and privacy.
The goal of this tutorial is threefold: First, we describe a generalized framework for biometric template protection - a single abstract construction consisting of an encoding module and a decision module that brings all major template protection architectures under a common umbrella. The framework captures the essential differences between traditional biometrics and biometric template protection. It allows us to define measures of performance (accuracy and robustness), security, privacy and storage. We highlight the fact that, while the accuracy/robustness (FAR vs. FRR) tradeoff is well understood, the security/privacy tradeoff has interesting non-obvious mathematical properties. Furthermore, we stress that the desire for increased security, privacy and template revocability imposes new challenges on biometric feature extraction: New methods are needed to prevent degradation of the accuracy/robustness tradeoffs that we take for granted in conventional biometric systems.
Second, we describe four major architectures for biometric template protection, casting them as realizations of the above generalized framework. These include signal processing-based architectures such as cancelable biometrics, cryptography-based architectures using multiparty computation, and information theory-based architectures such as fuzzy commitment and secure sketches. Example implementations of each architecture are discussed, including those based on warping transforms, random projections, homomorphic encryption, and error correcting codes. Our development first emphasizes the common themes underlying these architectures, and then draws attention to their differences.
Third, we draw attention to the numerous exciting challenges at the forefront of research in this field. These include analyses of practical situations where a single user has enrolled at multiple biometric devices. We highlight several important issues that will be of interest to practitioners and researchers alike, including emerging biometric modalities, the challenge of privacy- preserving biometric alignment, the need for standardization, consideration of malicious attacks, and the emergence of new mathematical tools for biometric encryption.