Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Generative Adversarial Network on Motion-Blur Image Restoration

Zhengdong Li

TL;DR

This work addresses motion blur in photographs by employing a GAN-based deblurring framework that accepts blurred inputs and generates sharp outputs, trained on a GoPro paired dataset with perceptual loss guidance. The Generator uses a ResNet-based architecture and is trained adversarially against a Discriminator, optimized with the minimax objective $\min_G \max_D V(D,G) = E_x[ \log D(x) ] + E_z[ \log (1 - D(G(z))) ]$ and a perceptual loss built on feature representations $\phi$, such as $\frac{1}{N} \sum_{i=1}^{N} \| \phi(y_{true}) - \phi(y_{pred}) \|^2$. Results show a mean PSNR of $29.1644$ and mean SSIM of $0.7459$, with an average processing time of $4.6921$ s per image, competitive with DeblurGAN baselines though room remains for SSIM gains due to dataset scale. The study demonstrates the viability of GAN-based motion blur restoration in practical settings and outlines concrete paths—larger datasets, GPU acceleration, and alternative backbones—to further enhance image fidelity and restoration speed for real-world applications such as smartphone photography.

Abstract

In everyday life, photographs taken with a camera often suffer from motion blur due to hand vibrations or sudden movements. This phenomenon can significantly detract from the quality of the images captured, making it an interesting challenge to develop a deep learning model that utilizes the principles of adversarial networks to restore clarity to these blurred pixels. In this project, we will focus on leveraging Generative Adversarial Networks (GANs) to effectively deblur images affected by motion blur. A GAN-based Tensorflow model is defined, training and evaluating by GoPro dataset which comprises paired street view images featuring both clear and blurred versions. This adversarial training process between Discriminator and Generator helps to produce increasingly realistic images over time. Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM) are the two evaluation metrics used to provide quantitative measures of image quality, allowing us to evaluate the effectiveness of the deblurring process. Mean PSNR in 29.1644 and mean SSIM in 0.7459 with average 4.6921 seconds deblurring time are achieved in this project. The blurry pixels are sharper in the output of GAN model shows a good image restoration effect in real world applications.

Generative Adversarial Network on Motion-Blur Image Restoration

TL;DR

This work addresses motion blur in photographs by employing a GAN-based deblurring framework that accepts blurred inputs and generates sharp outputs, trained on a GoPro paired dataset with perceptual loss guidance. The Generator uses a ResNet-based architecture and is trained adversarially against a Discriminator, optimized with the minimax objective and a perceptual loss built on feature representations , such as . Results show a mean PSNR of and mean SSIM of , with an average processing time of s per image, competitive with DeblurGAN baselines though room remains for SSIM gains due to dataset scale. The study demonstrates the viability of GAN-based motion blur restoration in practical settings and outlines concrete paths—larger datasets, GPU acceleration, and alternative backbones—to further enhance image fidelity and restoration speed for real-world applications such as smartphone photography.

Abstract

In everyday life, photographs taken with a camera often suffer from motion blur due to hand vibrations or sudden movements. This phenomenon can significantly detract from the quality of the images captured, making it an interesting challenge to develop a deep learning model that utilizes the principles of adversarial networks to restore clarity to these blurred pixels. In this project, we will focus on leveraging Generative Adversarial Networks (GANs) to effectively deblur images affected by motion blur. A GAN-based Tensorflow model is defined, training and evaluating by GoPro dataset which comprises paired street view images featuring both clear and blurred versions. This adversarial training process between Discriminator and Generator helps to produce increasingly realistic images over time. Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM) are the two evaluation metrics used to provide quantitative measures of image quality, allowing us to evaluate the effectiveness of the deblurring process. Mean PSNR in 29.1644 and mean SSIM in 0.7459 with average 4.6921 seconds deblurring time are achieved in this project. The blurry pixels are sharper in the output of GAN model shows a good image restoration effect in real world applications.
Paper Structure (9 sections, 2 equations, 6 figures, 4 tables)

This paper contains 9 sections, 2 equations, 6 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Proposed Methods
  4. Experiments
  5. Training Process
  6. GAN Architecture
  7. Results
  8. Future Work
  9. Conclusion

Figures (6)

  • Figure 1: The working flow for Generator and Discriminator in GAN with GoPro dataset
  • Figure 2: Graphical representation of the architecture of Generator
  • Figure 3: Graphical representation of the architecture of Discriminator
  • Figure 4: Sample output using blurry image from GoPro dataset in validation part.
  • Figure 5: Sample output with self-proposed blurry images.
  • ...and 1 more figures