Weihong Deng

Weihong Deng, Ph.D

Beijing University of Posts and Telecommunications, 
Homepage: www.whdeng.cn, E-mail: whdeng@bupt.edu.cn'

Weihong Deng received the B.E. degree in information engineering and the Ph.D. degree in signal and information processing from the Beijing University of Posts and Telecommunications (BUPT), Beijing, China, in 2004 and 2009, respectively. From Oct. 2007 to Dec. 2008, he was a postgraduate exchange student in the School of Information Technologies, University of Sydney, Australia. He is currently an associate professor in School of Information and Telecommunications Engineering, BUPT. His research interests include statistical pattern recognition and computer vision, with a particular emphasis in face recognition. He has published over 50 technical papers in international journals and conferences, such as IEEE TPAMI and CVPR. He serves as guest editor for Image and Vision Computing Journal and the reviewer for several international journals, such as IEEE TPAMI / TIP / TIFS / TNNLS / TMM / TSMC, IJCV, PR / PRL. Recently, he gives tutorials on face recognition at ICME 2014, ACCV 2014, CVPR2015 and FG2015, and organizes the workshop on feature and similarity learning in ACCV2014 with colleagues. His Dissertation titled “Highly accurate face recognition algorithms” was awarded the Outstanding Doctoral Dissertation Award by Beijing Municipal Commission of Education in 2011. He has been supported by the program for New Century Excellent Talents by the Ministry of Education of China in 2013 and Beijing Nova Program in 2016.

Talk I: Face Recognition: From Subspace to Feature Learning

Face recognition is a longstanding computer vision problem and a variety of methods have been proposed for face recognition over the past two decades. In this talk, I will first introduce the typical system pipeline, and historical algorithm landmarks, and recent progresses of face biometric. I also review face detection and alignment as two underestimated components that significantly affect the real-world performance of a face recognition system. Most importantly, I will overview the trend from traditional subspace learning to the emerging feature learning techniques and discuss how they are employed to boost face recognition performance. I briefly introduce the basic concept of subspace and feature learning, and show the key advantages and disadvantages of existing feature learning methods in different face recognition tasks. Lastly, I will discuss some open problems in feature learning to show how to further develop more advanced feature learning algorithms for real-world face recognition.

Talk II: Metric Learning for Face Recognition

In this talk, I will overview the trend of metric learning techniques and discuss how they advance various face recognition applications. This talk includes three parts. First, I will briefly introduce the basic concept of metric learning, and show how they are used to improve the performance of different face recognition tasks in previous work. Second, I will introduce several our newly proposed metric learning methods include cost-sensitive metric learning, sparse reconstruction metric learning, locality repulsed metric learning, and discriminative deep metric learning. Then, I 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, I will discuss some open problems to understand how to develop more advanced metric learning algorithms for human face analysis in the future.