报告摘要：In recent times, an increasing, worldwide effort has been devoted to the development of automatic personal identification systems that can be effective in a wide variety of security contexts. As one of the most powerful and reliable means of personal authentication, biometrics has been an area of particular interest. It has led to the extensive study of biometric technologies and the development of numerous algorithms, applications, and systems, which could be defined as Advanced Biometrics. This presentation will systematically explain this new research trend. As case studies, a new biometrics technology (palmprint recognition) and two new biometrics applications (medical biometrics and aesthetical biometrics) are introduced. Some useful achievements could be given to illustrate their effectiveness.

报告摘要：生物特征识别技术的广泛使用给百姓生活和社会治理带来了极大的方便，同时也产生了极大的安全和隐私隐患。近年来世界各国对人工智能技术特别是生物特征识别使用的安全、伦理、隐私等问题制定了一系列的法规。生物特征识别技术安全可靠的使用，一方面有赖于法律法规的健全，同时也和技术本身的进步以及合理使用有着极大的关系。如何防止身份假冒与盗用等安全和隐私问题，在技术上应该采取哪些措施加以防范是生物特征识别领域值得关注的问题。报告将从生物特征识别技术应用的历史变迁探讨其中的问题，同时将报告一些近年来在这方面研究的探索与进展。

报告摘要：Face recognition is possibly one of the most successful applications of Computer Vision and AI. Today's information technology allowed to deploy face recognition in several domains, ranging from automated border control to mobile device authentication. Even though the progress in computing power and machine learning allowed to implement very fast and efficient systems, there are still several issues which remain unsolved. On the other hand, the basic "face recognition pipeline", conceived 30 years ago, still remains unaltered. As such, we need to learn from the past and address some research questions which are still unanswered. Among them: 1. If face recognition is a "solved" problem, why are we still doing research on this topic? 2. What are the drawbacks and limitations of current deep learning models? How far can we go by exploiting increasing amounts of face data? 3. Is the human visual system still the best comparative face recognition model? If so, what can we learn from the way humans recognize faces? 4. How can we build "ethical" systems which properly address current privacy concerns? In this talk we'll address these questions, trying to envisage a path forward with the aim of driving our research curiosity towards the design of tomorrow's intelligent machines.

报告摘要：Physiological signals, including e.g., heart rate (HR), heart rate variability (HRV), and respiratory frequency (RF) are important indicators of our health, which are usually measured in clinical examinations. Traditional physiological signal measurement often involves contact sensors, which is inconvenient or cause discomfort in long-term monitoring sessions. This talk focuses on studies exploring remote HR measurement from facial videos, from early developed hand-crafted to later proposed deep learning based methods, together with a new Oulu Bio-Face (OBF) database as a benchmark dataset. Moreover, applications of remote heart rate measure are presented, including e.g., atrial fibrillation (AF) screening and face anti-spoofing, with the video-extracted heart rate related features.

报告摘要：Most existing biometrics are based on visual signals. In this talk, I discuss the possibility of using signals beyond the ones captured by traditional RGB cameras. This work is inspired by our latest efforts on constructing digital humans more than their appearances, by incorporating anatomical structures. For example, we have previously used XRay and MRI to capture the inner structures (bones, muscles, etc) of faces and hands as well as use IMU units to capture whole body movements. I show many of these signals can potentially serve as new types of biometrics and benefit applications where RGB signals fail (e.g., low light, long range, fast movements, etc).

报告摘要：以人为中心的计算是计算机视觉中的核心问题, 其核心的感知与识别人的表观、身份、动作、情感和意图以及根据相关信息生成真实的虚拟数字人。尽管人作为整体, 其外观、身份、动作、情感等是有机联系的整体，但是目前以人为中心的研究仍然以研究人的某个方面任务为主，例如行人重识别、行人检测、人体动作生成、人头特写等。这种研究范式未能挖掘以人为中心的任务之间的相关性, 不能充分挖掘现有公开数据的潜力。本次报告将介绍以人为中心的通用识别模型、高泛化能力的生物特征识别大模型以及多模态可控的人体动作生成大模型的构建中对于数据、架构与路径的探索，为迈向高精度、通用、对人可全面理解的识别和生成大模型铺路。