Peningkatan Kualitas Gambar Wajar Pada Sistem Deteksi Wajah Menggunakan Generative Face Prior-Generative Adversarial Network (GFPGAN)
DOI:
https://doi.org/10.34010/z6n9am93Abstract
This study investigates the capability of the Generative Adversarial Network (GAN) algorithm, specifically GFPGAN (Generative Face Prior-Generative Adversarial Network), in enhancing the quality of facial images to support more accurate face recognition. GFPGAN is known for its effectiveness in restoring degraded facial images, such as those affected by blurriness, noise, and low resolution—common issues in CCTV (Closed-Circuit Television) footage or other low-quality image sources. By leveraging the GAN architecture, which consists of a generator and a discriminator, GFPGAN is able to produce high-detail facial images while preserving the original facial identity. In this research, GFPGAN is utilized to restore degraded facial images prior to the face recognition process using the DLIB library. Various types of image degradation are tested, including blur, noise, grayscale conversion, and JPEG (Joint Photographic Experts Group) compression. The evaluation involves comparing the face recognition success rate using DLIB before and after the restoration process with GFPGAN. The results demonstrate that previously unrecognizable images become identifiable after being processed with GFPGAN, thereby confirming that image restoration can significantly improve face recognition accuracy.
References
[1] Li, Z., & Jain, A. K. Face recognition: Challenges and research directions. arXiv preprint arXiv:1905.01349. 2019
[2] Goodfellow, I., et al. Generative adversarial nets. Advances in Neural Information Processing Systems (NeurIPS). 2014
[3] Ledig, C., et al. Photo-realistic single image super-resolution using a generative adversarial network. Proc. IEEE CVPR. 2017
[4] Yang, Y., Zhang, X., Tian, X., & Fu, Y. GPEN: Generative Prior Engine for Blind Face Restoration. Proc. ICCV. 2021
[5] Wang, X., et al. GFPGAN: Towards Real-World Blind Face Restoration with Generative Facial Prior. arXiv preprint arXiv:2101.04061. 2021
[6] Johnson, J., Alahi, A., & Fei-Fei, L. Perceptual losses for real-time style transfer and super-resolution. Proc. ECCV. 2016
[7] Arjovsky, M., Chintala, S., & Bottou, L. Wasserstein GAN. arXiv preprint arXiv:1701.07875. 2017
[8] He, R., Wu, X., Sun, Z., & Tan, T. WiderFace and Face Recognition in the Wild: A Review. IEEE TPAMI, 44(10), 7006–7022. 2022
[9] Qiu, Z., Li, H., Wang, Y., & Guo, J. Blind face restoration for improved face recognition in surveillance. IEEE Access, 9, 123010–123021. 2021
[10] Zhang, R., et al. The Unreasonable Effectiveness of Deep Features as a Perceptual Metric, CVPR, 2018
[11] Heusel, M., et al. GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium, NeurIPS, 2017.
[12] Hore, A., and Ziou, D. Image Quality Metrics: PSNR vs. SSIM, ICPR, 2010.
[13] Wang, X., et al. Real-ESRGAN: Training Real-World Blind Super-Resolution with Pure Synthetic Data, ICCV Workshops, 2021.
[14] King, D.E. Dlib-ml: A Machine Learning Toolkit, Journal of Machine Learning Research, 2009.
[15] Schroff, F., Kalenichenko, D., & Philbin, J. FaceNet: A Unified Embedding for Face Recognition and Clustering, CVPR, 2015
[16] S. Sugeng and A. Mulyana, "Sistem Absensi Menggunakan Pengenalan Wajah (Face Recognition) Berbasis Web LAN," Jurnal Sisfokom (Sistem Informasi dan Komputer), vol. 11, no. 1, pp. 127–135, 2022
[17] Paszke, A., et al. PyTorch: An Imperative Style, High-Performance Deep Learning Library, NeurIPS, 2019.
