Title of the Talk: Recognizing Human Identity, Age, and Aesthetic Attributes from Gait
Speaker: Yasushi Makihara, Osaka University, Japan
Abstract: Gait, i.e., a way of walking, is considered as one of behavioral biometrics and contains a variety of information such as identity, age, gender, disease/health status, and aesthetic attribute. Particularly, gait-based person identification, i.e., gait recognition has been extensively studied for the last two decades because the gait can be recognized even at a distance from a camera without subjects' cooperation, unlike the other biometrics such as DNA, fingerprint, vein, and iris. The gait recognition is therefore expected to be applied to the criminal investigation, forensic science, and surveillance using CCTV footage. However, the absence of the subjects' cooperation may sometimes induce large intra-subject variations of the gait due to the changes of viewpoints, walking directions, speeds, clothes, and shoes. In this talk, I will introduce a recent progress of our studies on gait recognition robust against the above-mentioned covariates. Moreover, some other applications of the video-based gait analysis will be also introduced, including but not limited to age and gender estimation, medical or health status estimation, and aesthetic attribute estimation.
Title of the Talk: Global ID: Biometrics for Billions of Identities
Speaker: Sharath Pankanti, Microsoft, UDA
(work done with Anil K. Jain, Karthik Nandakumar, Salil Prabhakar, Sunpreet S. Arora, Anoop M. Namboodiri, and Arun Ross)
Abstract: The world’s population, which is currently estimated to be nearly eight Billion, is very likely to surpass 10 Billion by the turn of the century. While there are several challenges when dealing with a population of this magnitude, the ability to positively establish or refute an individual’s identity is likely to be one of the fundamental expectations of a global society. In this presentation, we systematically discuss the issues impacting the design, implementation and deployment of a large-scale biometric identification system that can effectively manage and distinguish over 10 Billion identities. In this regard, we identify four technological issues that have to be satisfactorily resolved to design such a system: system scalability, identification accuracy, response time, template security and privacy. We discuss how the lessons learned from ongoing large-scale biometric systems such as UAE’s Border Crossing System and India’s National ID Card Program (Aadhaar) can be leveraged and incorporated into a Global ID system that handles 10B identities. Further, we study existing large-scale pattern recognition and machine learning systems, and determine how the challenges resident in such systems can be effectively addressed for use in the proposed Global ID system. Finally, we assess the gaps that need to be addressed by the research and development community-at-large for designing the Global ID system. We conclude that the outstanding research, engineering and design topics are “Grand Challenges” and, without a serious understanding of the underlying complex issues, simplistic identity infrastructure solutions will be dwarfed by the enormity of the identity problems of the next generation.
Title of the Talk: Face Presentation Attack Detection
Speaker: PC Yuen, Hong Kong Baptist University, Hong Kong
Abstract: Face recognition technology has been successfully deployed in many practical applications in the past two decades, ranging from law enforcement to e-payment on mobile phones. While face recognition for person authentication offers great convenience, face biometrics system security is not fully addressed. In this talk, I will discuss the challenges of face presentation attack detection and briefly review the existing methods. After that, I will share our recent works on 3D mask presentation attack detection (PAD), generalized face PAD, and federated learning-based PAD. Finally, I will also discuss current research problems in PAD.