Face super-resolution in biometrics: Recent advances and future challenges
Face super-resolution (or face hallucination) refers to the task of recovering high-resolution facial images from corresponding low-resolution inputs. Solutions to this task have important applications in face-oriented vision problems, such as face editing, face alignment, 3D face reconstruction, face attribute estimation and most notably face recognition. Driven by advances in deep learning, recent years have seen tremendous progress in this area with contemporary deep face hallucination models achieving formidable performance. In my presentation, I will first talk about the recent progress in the field of face super-resolution, review existing approaches and discuss the most important trends in the area. Next, I will present our solution to the problem of face hallucination, which uses explicit identity constraints in addition to the common reconstruction loss during model training and show how it compares to state-of-the-art hallucination models from the literature. I will elaborate on the main limitations of existing face super-resolution approaches and present challenges that will need to be addressed in the future. Finally, I will describe our approach to face recognition from low-resolution images that relies on super-resolution and is shown to result in state-of-the-art performance on a popular benchmark.
Date and Time
Location
Hosts
Registration
- Date: 03 Dec 2020
- Time: 02:15 PM to 03:15 PM
- All times are (GMT+01:00) CET
-
Add Event to Calendar
- Contact Event Host
- Co-sponsored by Roma Tre University, Dept. of Engineering, Section of Applied Electronics
Speakers
Vitomir Struc of University of Ljubljana, Slovenia
Face super-resolution in biometrics: Recent advances and future challenges
Face super-resolution (or face hallucination) refers to the task of recovering high-resolution facial images from corresponding low-resolution inputs. Solutions to this task have important applications in face-oriented vision problems, such as face editing, face alignment, 3D face reconstruction, face attribute estimation and most notably face recognition. Driven by advances in deep learning, recent years have seen tremendous progress in this area with contemporary deep face hallucination models achieving formidable performance. In my presentation, I will first talk about the recent progress in the field of face super-resolution, review existing approaches and discuss the most important trends in the area. Next, I will present our solution to the problem of face hallucination, which uses explicit identity constraints in addition to the common reconstruction loss during model training and show how it compares to state-of-the-art hallucination models from the literature. I will elaborate on the main limitations of existing face super-resolution approaches and present challenges that will need to be addressed in the future. Finally, I will describe our approach to face recognition from low-resolution images that relies on super-resolution and is shown to result in state-of-the-art performance on a popular benchmark.
Biography:
Vitomir Štruc is an Associate Professor at the University of Ljubljana, Slovenia. He received his doctoral degree from the Faculty of Electrical Engineering in Ljubljana in 2010. Vitomir's research interests include problems related to biometrics, computer vision, image processing, pattern recognition and machine learning. He (co-)authored more than 100 research papers for leading international peer reviewed journals and conferences in these and related areas. He served in different capacities on the organizing committees of several top-tier vision conferences, including IEEE Face and Gesture, ICB, WACV and IJCB. Vitomir is a Senior Area Editor for the IEEE Transactions on Information Forensics and Security, and an Associate Editor for Pattern Recognition, Signal Processing, and IET Biometrics. He served as an Area Chair for WACV 2018, 2019, 2020, ICPR 2018, Eusipco 2019 and FG 2020. Dr. Struc is a member of the IEEE, IAPR, EURASIP, Slovenia’s national contact point for the EAB and the current president of the Slovenian Pattern Recognition Society, the Slovenian branch of IAPR
Email: