1. CSN
  2. All Volumes and Issues
  3. Volume 7, Number 1, 2025
  4. Evaluation of Deep Learning-based Super-resolution Methods for Enhanced Facial Identification Accuracy

Evaluation of Deep Learning-based Super-resolution Methods for Enhanced Facial Identification Accuracy

CSN.
2025;
Volume 7, Number 1
: pp. 295 - 306
https://doi.org/10.23939/csn2025.01.295
Authors:
  1. D. O. Khanin
  2. V. I. Otenko
1
Lviv Polytechnic National University, Department of Information Protection
2
Lviv Polytechnic National University, Department of Information Protection

This paper presents a comparative analysis of modern super-resolution (SR) methods for improving the accuracy of face recognition in video surveillance systems. The low quality of images obtained from surveillance cameras is a significant obstacle to effective person identification, making the use of SR methods particularly relevant.
Both classical interpolation methods (bicubic interpolation) and deep learning-based methods, including convolutional neural networks (SRCNN) and generative adversarial networks (ESRGAN, Real- ESRGAN, FSRNet), are analyzed. The methods were evaluated based on criteria such as accuracy (PSNR, SSIM), processing speed, computational resource requirements, and the availability of ready-made implementations. The study showed that deep learning-based methods significantly outperform traditional approaches in terms of reconstruction quality, especially in restoring fine details and textures important for face recognition.
It was determined that Real-ESRGAN is one of the most promising methods for practical application due to its ability to effectively process real-world low-quality images, while FSRNet offers an optimal balance between accuracy and speed for face recognition tasks. The choice of these methods as the most promising is justified, and directions for further research are outlined, including optimization of existing algorithms, adaptation to specific shooting conditions, and the development of comprehensive end-to-end face recognition systems based on SR. The results of the study emphasize the importance of using SR methods to improve the efficiency of security systems operating under conditions of limited image quality.

super-resolution
face recognition
deep learning
convolutional neural networks
Generative Adversarial Networks
video surveillance systems
PSNR
SSIM
  1. Dong C., Loy C. C., He K., Tang X. Image Super-Resolution Using Deep Convolutional Networks. IEEE Transactions     on     Pattern     Analysis     and     Machine     Intelligence.     2016.     Vol.     38.     295-307.     Doi: 10.1109/TPAMI.2015.2439281.
  2. Wang X., Yu K., Dong C., Loy C. C. ESRGAN: Enhanced Super-Resolution Generative Adversarial Networks, ECCV Workshops. 2018. URL: https://arxiv.org/abs/1809.00219 (Accessed: 17 January 2025).
  3. Aakerberg A., Nasrollahi K., Moeslund T. B. Real-World Super-Resolution of Face-Images from Surveillance Cameras, arXiv preprint arXiv:2102.03113. 2021. Available at: https://arxiv.org/abs/2102.03113 (Accessed: 17 January 2025).
  4. Singh N., Rathore S. S., Kumar S. Towards a super-resolution based approach for improved face recognition in low-resolution environment, Multimedia Tools and Applications. 2022. Doi: 10.1007/s11042-022- 13160-z (Accessed: 17 January 2025).
  5. Chen Y., Tai Y., Liu X. FSRNet: End-to-End Learning Face Super-Resolution with Facial Priors, CVPR. 2018. Available at: https://openaccess.thecvf.com/content_cvpr_2018/html/Chen_FSRNet_End-to-... CVPR_2018_paper.html.
  6. Wang X., Chan K. C., Yu K., Dong C., Loy C. C. Real-ESRGAN: Training Real-World Blind Super- Resolution with Pure Synthetic Data, ICCV Workshops. 2021. Available at: https://arxiv.org/abs/2107.10833 (Accessed: 17 January 2025).
  7. Schroff F., Kalenichenko D., Philbin J. FaceNet: A Unified Embedding for Face Recognition and Clustering, CVPR. 2015. Available at: https://arxiv.org/abs/1503.03832 (Accessed: 17 January 2025).
  8. Deng J., Guo J., Xue N., Zafeiriou S. ArcFace: Additive Angular Margin Loss for Deep Face Recognition, CVPR. 2019. Available at: https://arxiv.org/abs/1801.07698 (Accessed: 17 January 2025).
  9. Wang Z., Chen J., Hoi S. C. Deep Learning for Image Super-Resolution: A Survey/ IEEE Transactions on Pattern Analysis and Machine Intelligence. 2020. Vol. 43. 3365–3387. Doi: 10.1109/TPAMI.2020.2982166.
  10. Park S., Kim J., Lee J. Super-Resolution for Surveillance Images. Applied Sciences. 2021. Vol. 11. 942. Available at: (link) (Accessed: 17 January 2025).
  11. Wang X., Li J., Luo Z. Real-World Super-Resolution for Recognition Tasks. IEEE Access. 2022. Vol. 10. 63421–63430. Doi: 10.1109/ACCESS.2022.3185525.
  12. Kim S., Choi H., Jung K. Face Super-Resolution Using Deep Learning. Pattern Recognition Letters. 2023. Vol. 163. 53–61. Doi: 10.1016/j.patrec.2023.02.005.