Title: Innovative approach of masked facial recognition

Authors: Om Pradyumana Gupta; Arun Prakash Agrawal; Om Pal

Addresses: Department of Computer Science and Engineering, Sharda University, Greater Noida, India ' Department of Computer Science and Applications, Sharda University, Greater Noida, India ' Department of Computer Science, University of Delhi, New Delhi, India

Abstract: Not only diseases such as COVID-19 but also professions across healthcare, construction and manufacturing require the usage of facemasks. Currently, obstacles that are present in the available masked facial recognition techniques lack a focus on Indian specific databases which poses serious challenges in surveillance. In this paper, the proposed model, which achieved 96.7% accuracy, works in two-stages. The input goes through the pre-processing stage, which is based on an image cropping approach with a multi-task cascaded convolutional neural network applied over it. Output moves to the convolutional neural network stage where the image is trained on two models, convolutional block attention module and modified face net architecture. Special techniques such as channel attention, spatial attention, batch normalisation, leaky rectification and face embedding vector normalisation, triplet selection are applied to reach an optimal level of accuracy. The robotic process automation engine using Bayesian optimisation is also put in place.

Keywords: image processing; face recognition; artificial intelligence; Python; masked face detection; deep learning; neural network; COVID-19; facial landmark points; feature extraction.

DOI: 10.1504/EG.2025.143118

Electronic Government, an International Journal, 2025 Vol.21 No.1, pp.35 - 67

Received: 19 May 2023
Accepted: 12 Mar 2024
Published online: 03 Dec 2024 *

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