Towards efficient and robust face recognition through attention-integrated multi-level CNN
RobFaceNet is a robust and efficient CNN designed for face recognition, achieving a balance between high accuracy and computational efficiency by using multiple features and attention mechanisms to enhance facial feature extraction. It outperforms state-of-the-art models like ArcFace on recognition tasks while being significantly more lightweight, making it suitable for resource-constrained devices like robots, embedded systems, and mobile devices..
Towards efficient and robust face recognition through attention-integrated multi-level CNN
The rapid development of deep Convolutional Neural Networks (CNNs) has significantly advanced face verification technology, but their complexity and high computational demands make them challenging to implement on resource-limited devices. To address this, RobFaceNet is introduced as a robust and efficient CNN specifically for face recognition. RobFaceNet balances accuracy and computational efficiency by utilizing multiple features and attention mechanisms, including both low- and high-level attributes and a novel bottleneck integrating channel and spatial attention. This design improves facial feature extraction and robustness. Experimental results show that RobFaceNet outperforms state-of-the-art models like ArcFace on FR datasets (e.g., 95.95% vs. 95.45% on CA-LFW) while being more lightweight, using only 3% of the parameters and significantly fewer FLOPs (337M vs. 24211M). This makes RobFaceNet ideal for use on robots, embedded systems, and mobile devices.
Fulltext Access
https://link.springer.com/article/10.1007/s11042-024-19521-0
Citing
@article{khalifa2024towards,
title={Towards efficient and robust face recognition through attention-integrated multi-level CNN},
author={Khalifa, Aly and Abdelrahman, Ahmed A and Hempel, Thorsten and Al-Hamadi, Ayoub},
journal={Multimedia Tools and Applications},
pages={1--23},
year={2024},
publisher={Springer}
}