Enhancing Image Deblurring with Advanced Learning

Abstract

This study presents a novel framework for adaptive photo restoration by integrating GFPGAN for face-focused enhancement and Real-ESRGAN for general image refinement. Users select modes tailored to image content. The model features blur-aware preprocessing, intelligent background boosting, and output evaluation through SSIM. The application is deployed using a user-friendly Gradio interface and shows consistent performance across varied visuals.

Introduction

This project introduces a dual-mode image enhancement application that leverages recent advancements in neural networks to restore degraded visuals with improved clarity. Unlike traditional models limited to fixed use cases, this system intelligently adapts enhancement methods based on image content—distinguishing between faces and natural scenes.

Key Contributions

• Problem Addressed: Traditional image enhancement models lack adaptability and struggle with varying image types and degradation levels.

• Proposed Solution: A two-mode enhancement system that routes images to specialized models:

  • GFPGAN for facial image restoration

  • Real-ESRGAN for landscape and general scene enhancement

• Preprocessing Logic: A smart filtering mechanism analyzes image blur and clarity before enhancement to guide optimal model selection.

Literature Insights

• Wang et al. (2021): Developed Real-ESRGAN for natural scene reconstruction.

• GFPGAN: Known for handling facial image degradation effectively.

• Lim et al. and Isola et al.: Contributed to residual learning and image-to-image translation techniques.

System Workflow

1. Image Upload

2. Mode Selection (Face or Scene)

3. Smart Preprocessing (blur detection, contrast tuning)

4. Model Routing (GFPGAN or Real-ESRGAN)

5. Optional Masking (to isolate enhancement regions)

6. Result Generation

7. Quality Evaluation (via SSIM – Structural Similarity Index)

Technologies Used

• Python

• OpenCV, NumPy, Gradio

• Pretrained GAN models: GFPGAN v1.3, Real-ESRGAN Anime x4

Results

• Significant improvements in image detail and clarity

• Side-by-side comparisons of original vs enhanced images show clear restoration

• SSIM scores and user feedback confirm visual enhancement effectiveness

Conclusion

A modular photo enhancement system was created by combining two specialized Generative AI models. This platform achieves accurate and high-quality restoration for various image types.

References

[1] Wang et al., “Real-ESRGAN”: Training Real- World Blind Super-Resolution with Pure Synthetic Data, 2021 [2] Wang et al., \"GFPGAN”: Towards Real- World Blind Face Restoration with Generative Facial Prior, 2021 [3] Lim et al., \"EDSR for Super-Resolution\" : EDSR improves upon traditional residual networks, 2017 [4] Isola et al., \"Image-to-Image”: Image-to- Image Translation with Conditional Adversarial Networks, 2017

Copyright

Copyright © 2025 Mrs. Md. Jareena Begum, T. Kowsalya, P. Srividya, P. Harika, P. Tejomai Siva Naga Priya. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.