Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

StrCGAN: A Generative Framework for Stellar Image Restoration

Shantanusinh Parmar, Silas Janke

TL;DR

StrCGAN tackles the challenge of restoring high-fidelity stellar imagery from low-resolution, ground-based observations by integrating multi-spectral fusion and astrophysical regularization into a CycleGAN-inspired framework. The architecture employs a three-layer encoder–decoder generator with attention and skip connections, along with a discriminator, trained on MobilTelesco data augmented with ground-truth references from all-sky surveys. Despite qualitative improvements in stellar morphology and sharper details, StrCGAN lags behind state-of-the-art diffusion models and other baselines on standard metrics (e.g., PSNR, SSIM, FID) due to data scarcity, training instability, and strong domain priors. The work highlights the need for domain-specific evaluation metrics and proposes future directions, including hybrid diffusion–GAN approaches and learned perceptual losses, to achieve both perceptual excellence and quantitative rigor in astronomical image restoration.

Abstract

We introduce StrCGAN (Stellar Cyclic GAN), a generative model designed to enhance low-resolution astrophotography images. Our goal is to reconstruct high fidelity ground truth like representations of stellar objects, a task that is challenging due to the limited resolution and quality of small-telescope observations such as the MobilTelesco dataset. Traditional models such as CycleGAN provide a foundation for image to image translation but often distort the morphology of stars and produce barely resembling images. To overcome these limitations, we extend the CycleGAN framework with some key innovations: multi-spectral fusion to align optical and near infrared (NIR) domains, and astrophysical regularization modules to preserve stellar morphology. Ground truth references from multi-mission all sky surveys spanning optical to NIR guide the training process, ensuring that reconstructions remain consistent across spectral bands. Together, these components allow StrCGAN to generate reconstructions that are visually sharper outperforming standard GAN models in the task of astrophysical image enhancement.

StrCGAN: A Generative Framework for Stellar Image Restoration

TL;DR

StrCGAN tackles the challenge of restoring high-fidelity stellar imagery from low-resolution, ground-based observations by integrating multi-spectral fusion and astrophysical regularization into a CycleGAN-inspired framework. The architecture employs a three-layer encoder–decoder generator with attention and skip connections, along with a discriminator, trained on MobilTelesco data augmented with ground-truth references from all-sky surveys. Despite qualitative improvements in stellar morphology and sharper details, StrCGAN lags behind state-of-the-art diffusion models and other baselines on standard metrics (e.g., PSNR, SSIM, FID) due to data scarcity, training instability, and strong domain priors. The work highlights the need for domain-specific evaluation metrics and proposes future directions, including hybrid diffusion–GAN approaches and learned perceptual losses, to achieve both perceptual excellence and quantitative rigor in astronomical image restoration.

Abstract

We introduce StrCGAN (Stellar Cyclic GAN), a generative model designed to enhance low-resolution astrophotography images. Our goal is to reconstruct high fidelity ground truth like representations of stellar objects, a task that is challenging due to the limited resolution and quality of small-telescope observations such as the MobilTelesco dataset. Traditional models such as CycleGAN provide a foundation for image to image translation but often distort the morphology of stars and produce barely resembling images. To overcome these limitations, we extend the CycleGAN framework with some key innovations: multi-spectral fusion to align optical and near infrared (NIR) domains, and astrophysical regularization modules to preserve stellar morphology. Ground truth references from multi-mission all sky surveys spanning optical to NIR guide the training process, ensuring that reconstructions remain consistent across spectral bands. Together, these components allow StrCGAN to generate reconstructions that are visually sharper outperforming standard GAN models in the task of astrophysical image enhancement.

Paper Structure

This paper contains 23 sections, 1 equation, 8 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Traditional Super-Resolution Techniques
  4. GAN-Based Image Enhancement
  5. Astronomical Image Processing with Deep Learning
  6. Feature Efficiency and Sparse Data Handling
  7. Methodology
  8. Dataset Construction
  9. Preprocessing Pipeline
  10. Ground-Truth Augmentation
  11. MobilTelesco Image Cropping and Normalization
  12. StrCGAN Architecture
  13. Generator
  14. Discriminator
  15. Experimentation
  16. ...and 8 more sections

Figures (8)

  • Figure 1: All targets from MobilTelesco Dataset. From Left: Aldebaran, Bellatrix, Betelgeuse, Elnath, Hassaleh, Pleiades, Zeta Tauri. Top: Ground Truths, Bottom MobilTelesco sets
  • Figure 2: Overview of the proposed StrCGAN framework
  • Figure 3: Loss curve for the StrCGAN Generator over 100 epochs in Training mode.
  • Figure 4: Inference comparision for Aldebaran from Left: Ground truth, MobilTelesco, StrCGAN generated image
  • Figure 5: Inference comparision for Betelgeuse from Left: Ground truth, MobilTelesco, StrCGAN generated image
  • ...and 3 more figures