TY - GEN
T1 - Reviving Fine Details in Historical Photographs Using A-ESRGAN
T2 - 12th International Conference on Intelligent Computing and Information Systems, ICICIS 2025
AU - Jegede, Yvonne C.
AU - Jegede, Temitope D.
AU - Shalan, Atef Mohamed
N1 - Publisher Copyright:
© 2025 IEEE.
PY - 2025
Y1 - 2025
N2 - Restoring degraded historical photographs poses challenges, including loss of fine-grained detail, limited model transparency, and inadequate evaluation frameworks. This study presents an enhanced deep learning pipeline, Attention-Enhanced Super-Resolution Generative Adversarial Networks (A-ESRGAN), to address these gaps. By integrating attention mechanisms into the RRDBNet generator within the Real-ESRGAN framework, the model focuses on semantically meaningful regions, such as facial features, inscriptions, and textures, thereby improving perceptual and structural fidelity. We synthetically degraded the curated public-domain dataset with Gaussian blur and JPEG compression to mimic real-world conditions. We evaluated the model on paired low- and high-resolution images using PSNR and frequency-spectrum analysis. Results show that A-ESRGAN enhances PSNR and restores high-frequency details critical for visual clarity and historical interpretation. This research offers a scalable, interpretable framework for ethical and transparent AI-driven digital heritage restoration.
AB - Restoring degraded historical photographs poses challenges, including loss of fine-grained detail, limited model transparency, and inadequate evaluation frameworks. This study presents an enhanced deep learning pipeline, Attention-Enhanced Super-Resolution Generative Adversarial Networks (A-ESRGAN), to address these gaps. By integrating attention mechanisms into the RRDBNet generator within the Real-ESRGAN framework, the model focuses on semantically meaningful regions, such as facial features, inscriptions, and textures, thereby improving perceptual and structural fidelity. We synthetically degraded the curated public-domain dataset with Gaussian blur and JPEG compression to mimic real-world conditions. We evaluated the model on paired low- and high-resolution images using PSNR and frequency-spectrum analysis. Results show that A-ESRGAN enhances PSNR and restores high-frequency details critical for visual clarity and historical interpretation. This research offers a scalable, interpretable framework for ethical and transparent AI-driven digital heritage restoration.
KW - A-ESRGAN
KW - Attention Mechanisms
KW - Digital Heritage Preservation
KW - Frequency Spectrum Analysis
KW - PSNR
KW - RRDBNet
KW - Super Resolution
UR - https://www.scopus.com/pages/publications/105031771387
U2 - 10.1109/ICICIS66182.2025.11313137
DO - 10.1109/ICICIS66182.2025.11313137
M3 - Conference article
AN - SCOPUS:105031771387
T3 - Proceedings - 2025 IEEE 12th International Conference on Intelligent Computing and Information Systems, ICICIS 2025
SP - 150
EP - 157
BT - Proceedings - 2025 IEEE 12th International Conference on Intelligent Computing and Information Systems, ICICIS 2025
PB - Institute of Electrical and Electronics Engineers Inc.
Y2 - 25 November 2025 through 26 November 2025
ER -