Title: CEVAB: NIR-VIS face recognition using convolutional encoder-based visual attention block

Authors: Patil Jayashree Madhukar; P.M. Ashok Kumar; R. Anitha

Addresses: Department of Computer Engineering, Dr. D.Y. Patil School of Engineering, Lohegaon, Pune, Maharashtra, 412105, India ' Department of Computer Science and Engineering, Koneru Lakshmaiah Education Foundation, Vaddeswaram, AP, 522302, India ' Department of Computer Science and Engineering, Koneru Lakshmaiah Education Foundation, Vaddeswaram, AP, 522302, India

Abstract: Recent research in night vision face recognition has spiked due to the rise of night-time surveillance in public areas, where cameras often use near infrared (NIR) images. This paper presents a new face recognition method, the convolutional encoder-based visual attention block (CEVAB), optimised for NIR and visible spectrum (VIS) images. CEVAB combines a convolutional encoder with an attention-based architecture, focusing on critical facial features to enhance accuracy against watchlists. Tested on the FaceSurv dataset with over 132,000 images, CEVAB outshines traditional methods in VIS, achieving 95.08% Rank 1 accuracy at close distances, and in NIR, with 74.00% Rank 1 accuracy, surpassing competitors like Verilook and ResNet-50. These results prove CEVAB's exceptional adaptability and performance in various imaging conditions, significantly advancing night vision face recognition technology.

Keywords: deep learning; face recognition; NIR images; visual attention; convolutional decoder; convolutional encoder; cross-spectral recognition; deep learning; face recognition; feature extraction; night vision; NIR-VIS; surveillance systems; visual attention.

DOI: 10.1504/IJDATS.2024.140650

International Journal of Data Analysis Techniques and Strategies, 2024 Vol.16 No.3, pp.262 - 281

Received: 23 Jun 2023
Accepted: 26 Jan 2024
Published online: 29 Aug 2024 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article