Computer science team awarded for research on face recognition

Facial recognition matching process and the resulting scores for the matched images.

This novel method improves face recognition performance by choosing a suitable set of images from the database that match some critical properties of the query image (such as matching profile image against profile images or frontal against frontal). Figure on the right shows higher matching scores for the image pairs with the same property.

By Nicole Capozziello

Published March 2, 2022

A research team from the University at Buffalo’s Department of Computer Science and Engineering received the NVIDIA CCS Best Student Paper Runner Up Award at the Institute of Electrical and Electronics Engineers (IEEE) International Conference on Automatic Face and Gesture Recognition. 

Print
“This will help to improve trust in the use of face recognition technologies and improved verification capabilities in applications such as airport security.”
Nishant Sankaran, PhD graduate
Department of Computer Science and Engineering
Nishant Sankaran and Deen Dayal Mohan.

Nishant Sankaran and Deen Dayal Mohan

The premier international forum for research in image and video-based face, gesture and body movement recognition, the conference was held December 15-18, 2021.

The paper, TADPool: Target Adaptive Pooling for Set Based Face Recognition, describes an innovative method for improving face recognition performance when dealing with sets of face images captured under unconstrained settings, such as by surveillance cameras. This technology is particularly useful in law enforcement and homeland security scenarios where matching is done against sets of images captured under different conditions

“This will help to improve trust in the use of face recognition technologies and improved verification capabilities in applications such as airport security,” says Nishant Sankaran (PhD ’21, computer science and engineering), one of the paper’s first authors and a current applied scientist at Amazon.

“Ultimately, improvements in face recognition performance lead to fewer false-positive matches to subjects,” says Deen Dayal Mohan, the paper’s other first author and a PhD candidate in computer science and engineering. 

In addition to Mohan and Sankaran, the co-authors are: Sergey Tulyakov, a research scientist at UB’s Center for Unified Biometrics and Sensors (CUBS); Srirangaraj Setlur, principal research scientist and co-director of CUBS; and Venu Govindaraju, SUNY Distinguished Professor and founding director of CUBS.

CUBS’ research aims to advance the frontiers of biometric technologies across multiple traditional modalities such as face, fingerprint, and iris as well as novel biometrics, soft biometrics and multi-biometrics fusion. The work presented in this paper builds on prior funded research from IARPA (Intelligence Advanced Research Projects Activity, Office of the Director of National Intelligence) on the JANUS program for unconstrained face recognition.