Abstract:

Fingerprint/palmprint recognition is one of the important tools used in forensics to establish the identity of a suspect. Although latent print evidence has been widely used in forensics for more than a century, latent print identification continues to be a semi-automated process with significant human involvement in the feature extraction and matching stages. To reduce the workload of latent examiners, there is a critical and urgent need for fully automated latent identification systems. This tutorial will provide an introduction to recent advances in various topics in latent print recognition, including image enhancement, distortion rectification, matching, and indexing.

Biography:

Jianjiang Feng is an associate professor in Department of Automation, Tsinghua University, Beijing. He received his PhD degree from Beijing University of Posts & Telecommunications. After that, he has worked as a postdoc in the PRIP lab at Michigan State University. He has published over 30 papers, among which two received the best student paper awards on SIBGRAPI 2010 and IJCB 2011, and one received the best paper award on ICIEA 2012. He has served as program committee member of major biometrics conferences (ICB 2011-2014, BTAS 2012, SPIE Biometrics 2010-2011, CCBR 2012-2014), and has been a reviewer of several major image processing and pattern recognition journals (IEEE Trans. on TPAMI, TIP, TIFS). He has given tutorial talks on fingerprint recognition at several conferences (ICB 2013, CCBR 2013, Biometric Winter School 2014).

Abstract:

The past two decades have witnessed a considerable and rapid progress in human face analysis and a variety of face analysis methods have been proposed in the literature. Most existing face analysis methods have achieved reasonably good performance, especially in controlled conditions. However, in many real world applications, face images are usually captured in the wild and most existing face recognition methods are still far from satisfying. Since real‐life face images are usually affected by expressions, poses, occlusions, and illuminations, the difference of face images from the same class could be even larger than that from different classes. Therefore, how to learn a robust and discriminative distance metric to better measure the similarity of face samples that push the intra‐class variation and enlarge the inter‐class variation is a critical and challenging problem in face analysis. In recent years, many solutions have been proposed in this direction and significant improvements over the state‐of‐the‐arts have also been obtained by various metric learning methods.

In this tutorial, we will overview the trend of metric learning techniques and discuss how they advance different face analysis tasks such as face recognition, face verification, facial age estimation, kinship verification, and head pose estimation. This tutorial includes three parts. First, we briefly introduce the basic concept of metric learning, and show how they are used to improve the performance of different face analysis tasks in previous work. Second, we introduce several our newly proposed metric learning methods from two aspects: single‐metric learning and multi‐metric learning. For single‐metric learning, our methods include cost‐sensitive metric learning, sparse reconstruction metric learning, ordinal preserving metric learning, locality repulsed metric learning, and discriminative deep metric learning. For multi‐metric learning, we will present discriminative multi‐metric learning, multi‐view neighbourhood repulsed metric learning, localized multi‐kernel metric learning, and multi‐manifold metric learning. Third, we will present how these proposed metric learning methods are used to improve different face analysis tasks including face identification, face verification, video‐based face recognition, single‐sample face recognition, facial age estimation, kinship verification, and head pose estimation. Lastly, we will discuss some open problems to understand how to develop more advanced metric learning algorithms for human face analysis in the future.

Biography:

Jiwen Lu received the B.E. degree in mechanical engineering and the M.E. degree in electrical engineering, both from the Xi’an University of Technology, China, in 2003 and 2006, respectively, and the PhD degree in electrical engineering from the Nanyang Technological University, Singapore, in 2011. He is currently a research scientist at the Advanced Digital Sciences Center (ADSC), Singapore. His research interests include computer vision, pattern recognition, machine learning, biometrics, and multimedia. He has authored over 80 scientific papers in these areas, where 18 papers in IEEE Journals: TPAMI (3), TIP (1), TIFS (6), TCSVT (1), TSMC (2), and SPL (5), and 6 papers in top tier computer vision conferences: CVPR (3) and ICCV (3). He served as a technical program committee member for over 20 international conferences such as IEEE ICCV, CVPR, ECCV, ICPR, ICASSP, IJCB, and ICME, and a reviewer for over 40 international journals such as IEEE TPAMI, TIP, TIFS, TCSVT, TMM, TNNLS, TMM, and TSMC‐B. He co‐organizes several international workshops/competitions at some leading international conferences including ICME2014, ACCV2014, IJCB2014 and FG2015. He was a recipient of the First‐Prize National Scholarship and the National Outstanding Student Award from the Ministry of Education of China in 2002 and 2003, the 2012 Best Student Paper Award from PREMIA of Singapore, and the Best Paper Award Nominations from ICME2011 and ICME 2013, respectively. He gives a tutorial at IEEE ICME 2014.

Abstract:

This tutorial is intended for researchers who may be familiar with some other area of biometrics, and want an overview of the basic principles of iris recognition, history and current applications, and an introduction to some current major research issues. List of topics to be presented include: basic concepts of iris recognition; historical development; successful applications; effects of pupil dilation; effects of contact lenses; effects of iris aging.

Biography:

Kevin Bowyer is the Schubmehl-Prein Professor and Chair of the Department of Computer Science and Engineering at the University of Notre Dame. Professor Bowyer's research interests touch on various aspects of computer vision and pattern recognition, including biometrics and data mining. He is a Fellow of the IAPR, a Fellow of the IEEE, a Golden Core member of the IEEE Computer Society and received a 2014 IEEE Computer Society Technical Achievement Award "for pioneering contributions to the science and engineering of biometrics".

Professor Bowyer is serving as General Chair of the 2015 IEEE International Conference on Automated Face and Gesture Recognition. He was previously General Chair of the 2011 International Joint Conference on Biometrics, Program Chair of the 2011 Automated Face and Gesture Recognition conference, a founding General Chair of the IEEE Biometrics Theory Applications and Systems conference series, and a past EIC of the IEEE Transactions on Pattern Analysis and Machine Intelligence and the IEEE Biometrics Compendium. Professor Bowyer is a member of the Editorial Board for IEEE Access, IEEE's first "rapid publication, open access, maga-journal".

Abstract:

One of the major sources of variability in assessing the performance of a biometric system is the subject variability. Testing the same system on two disjoint populations of users almost always invariably yields two different results. This phenomenon was first described by Doddington et al when they evaluated speaker recognition systems in 1998. The users, or more precisely the statistical models associated with them, who are difficult to be recognised by a biometric system are given names such as goats and lambs; whereas users whose biometric traits are likely to be successful at impersonating others are called wolves. This phenomenon is dubbed “Doddington Zoo” and “Biometric Menagerie”. Subsequent studies then either aim at characterising the phenomenon, or at reducing the phenomenon that leads to better tailored decisions such as user-specific decision threshold and user-specific score calibration, and fusion strategies.

In this tutorial, we will describe biometric menagerie and explain how and why it has a direct impact on how the system performance is characterised; how confidence intervals can be estimated; and why performance prediction is difficult.

The significance of biometric menagerie is that it has impact on all biometric modalities. Furthermore, by reducing the phenomenon through user-specific strategies, a relative performance gain of about 30% has been observed for a unimodal biometric system; and up to 50% for a multimodal biometric system.

Biography:

Norman Poh joined the Department of Computing, University of Surrey, as a Lecturer in Multimedia Security and Pattern Recognition in 2012. He received the PhD degree from the Swiss Federal Institute of Technology in Lausanne (EPFL) entitled “Multi-system Biometrics: Optimal Fusion and User-specific Information”. His research interests focus on developing and applying pattern recognition theories to biometrics, information fusion, and healthcare informatics.

He received two personal Fellowships from the Swiss National Science Foundation (Young Prospective and Advanced Researcher grants) and received five best paper awards on works related biometrics (AVBPA’05, ICB’09, HSI 2010, ICPR 2010 and Pattern Recognition Journal 2006). He won the Researcher of the Year 2011 Award, University of Surrey. He is an Associate Editor of IET Biometrics Journal, an IEEE Certified Biometrics Professional and trainer, a member of IEEE and IAPR, and a member of the Education Committee of the IEEE Biometric Council.

He has proposed a number of strategies that characterise Biometric Menagerie as well as reducing the Menagerie effect. Examples are Biometric Menagerie Index, F-ratio normalisation (F-Norm), Model-specific Log-Likelihood Ratio (MS-LLR) Norm, Biometric-ratio (B-ratio), User-ranking, Discriminative normalisation, and Group-based normalisation. Other related works include quality-based information fusion and two-level bootstraps for confidence estimation.