Machine Learning for Biometrics

Prof. Dong Xu, University of Sydney, Australia (IEEE SPS Distinguished Lecturer)

Abstract: Over the past few decades, a large number of machine learning approaches including both dimension reduction methods and distance metric learning algorithms have been proposed for biometrics applications. In this talk, I will describe a general formulation known as graph embedding, which can unify a large family of dimension reduction algorithms within a single framework. This framework can also be used as a general platform for developing new dimension reduction algorithms. I will then introduce a series of new dimensionality reduction algorithms developed under this framework, in which image objects are represented in their intrinsic forms and orders (e.g., an image is a second-order tensor (i.e., a matrix) and sequential data such as video sequences used in event analysis is in the form of a third-order tensor). Finally, I will also introduce our recent distance metric learning works for biometrics applications.

Bio: Professor Dong Xu is Chair in Computer Engineering and ARC Future Fellow at the School of Electrical and Information Engineering, The University of Sydney, Australia. He received the B.Eng. and PhD degrees from University of Science and Technology of China, in 2001 and 2005, respectively.  Before joining The University of Sydney, he worked as a postdoctoral research scientist at Columbia University (2006-2007) and a faculty member at Nanyang Technological University (2007-2015).

Prof. Xu is an active researcher in the areas of image and video processing, computer vision and multimedia. He was selected as the Clarivate Analytics Highly Cited Researcher in the field of Engineering in 2018 and awarded the IEEE Computational Intelligence Society Outstanding Early Career Award in 2017. He was also selected to serve as an IEEE Signal Processing Society Distinguished Lecturer (2021-2022). Prof. Xu has published more than 150 papers in leading journals and conferences, among which two of his co-authored works (with his former PhD students) won the IEEE T-MM Prize Paper Award in 2014 and the CVPR Best Student Paper Award in 2010. According to Google Scholar, his publications have received over 20,000 citations.

Prof. Xu is/was on the editorial boards of ACM Computing Surveys, IEEE T-IP, T-PAMI, T-NNLS, T-MM and T-CSVT as well as other five journals, and he is serving/served as a guest editor of more than ten special issues in IJCV, T-NNLS, T-CSVT, T-CYB, IEEE Multimedia, ACM TOMM, CVIU and other journals. He is serving/served as a Program Chair of three international conferences including MLSP 2021, ICME 2014 and PCM 2012. He is also involved in the organization committees of many international conferences such as ACM MM 2021, GlobalSIP 2019, MMSP 2019, ICIP 2017, MMSP 2016 and VCIP 2015. He served as a steering committee member of ICME (2016-2017) and a track chair of ICPR 2016 as well as an area chair of AAAI 2020, ICCV 2017, ACM MM 2017, ECCV 2016 and CVPR 2012. He received the Best Associate Editor Award of T-CSVT in 2017. He is a Fellow of the IEEE and IAPR.

Image Provenance Analysis at Scale

Prof. Walter J. Scheirer, University of Notre Dame, USA

Abstract: Alarmingly, not all of the information available on the Internet is what it appears to be. It is now routine to encounter visual data that have been intentionally manipulated toward some business, political, or nihilistic end. Deceptive memes, bogus ads, and fabricated infographics are proliferating, with all threatening to undermine the public’s trust in information. Given the vast scale of the problem, an automated capability that can identify new instances of visual disinformation, trace its origin, and ultimately flag it as being problematic is needed. But compared to text, visual content presents unique challenges for media forensics. This talk presents an end-to-end processing pipeline for image provenance analysis, which works at real-world scale. It employs a cutting-edge image filtering solution that is able to find related images, as well as novel techniques for obtaining a provenance graph that expresses how the images, as nodes, are ancestrally connected. Building from provenance analysis, the talk goes on to introduce a scalable automated visual recognition pipeline for discovering meme genres of diverse appearance. This pipeline can ingest meme images from a social network, apply computer vision-based techniques to extract features and index new images into a database, and then organize the memes into related genres. Recent examples of visual disinformation will be highlighted, including repurposed imagery, parasitic advertising, and election-related memes. Finally, the talk will conclude with thoughts on continued research in this direction.

Bio: Walter J. Scheirer (Senior Member, IEEE) received the M.S. degree in computer science from Lehigh University, USA, in 2006, and the Ph.D. degree in engineering from the University of Colorado, Colorado Springs, CO, USA, in 2009. He is currently an Associate Professor with the Departmentof Computer Science and Engineering, University of Notre Dame, USA. Prior to that, he was a Post-Doctoral Fellow with Harvard University, USA, and the Director of Research and Development with Securics, Inc., from 2007 to 2012. His research interests include computer vision, machine learning, biometrics, and digital humanities.