Sparse Representation based Satellite Image Restoration Using Adaptive Reciprocal Cell

He Yanfei, Jianwei Zhang, Shunfeng Wang, Yuhui Zheng, Jin Wang, Yunjie ChenInternational Journal of Multimedia and Ubiquitous Engineering 2014 · DOI

Abstract

Recently, an emerging method called image sparse representation has attracted more attention. The method has been proved to be effective in various image processing applications. It is important to note that few sparse representation methods fail to analyze the aliasing in satellite image restoration. To address the problem, firstly, we employ adaptive reciprocal cell as an image quality estimation tool, which can analyze the satellite image degradation factors including aliasing, blur and noise. Then, with the help of this powerful tool, the estimation about the satellite image quality is introduced into the sparse representation model. Experiment results show that our method can produce good quality restored results.

Introduction

Satellite imagery is essential for various earth observation tasks, yet it is frequently plagued by three main types of degradation: blur (from the optical system), additive noise, and aliasing (caused by under-sampling at the sensor level). While many restoration techniques focus on deblurring and denoising, aliasing is often overlooked.

Standard approaches like Total Variation (TV) regularization are effective at removing noise but tend to over-smooth the image, destroying fine structural details. Emerging sparse representation (SR) methods—which represent image patches as a combination of "atoms" from a dictionary—often fail to account for the specific frequency-domain characteristics of aliasing in satellite sensors. This paper bridges this gap by creating a model that incorporates the sampling limits of the imaging system.

Methodology

The proposed method integrates Adaptive Reciprocal Cell (ARC) analysis with a modified sparse representation model.

Adaptive Reciprocal Cell (ARC) Analysis

The ARC is a tool derived from signal sampling theory that describes the "effective resolution" of an image acquisition system in the Fourier domain. By calculating the ARC, we define a support region ($\Omega_{arc}$) that identifies where the signal is reliable and where aliasing is likely to occur.

Modified Sparse Representation Model

Traditional SR models minimize the difference between the observed image and the reconstructed version in the spatial domain. This research moves the "data-fitting" term into the Fourier domain.

The innovation lies in restricting the data-fitting process to the support region defined by the ARC. Instead of a standard spatial error term, the model solves for a sparse coefficient vector $\alpha$ that minimizes: $$\|\mathcal{F}(f^p) - \mathcal{F}(h * \phi\alpha)\|_{2, \Omega_{arc}}^2 + \beta\|\alpha\|_1$$ Where $\mathcal{F}$ is the Fourier Transform, $h$ is the point-spread function, and $\phi$ is the dictionary. This ensures the restoration is guided by the most accurate frequency information available, effectively treating aliasing as a distinct sampling problem.

Experimental Setup

The study evaluates the proposed method using two primary sources:

  • Simulated Data: A high-quality image was artificially degraded using a Gaussian blur kernel ($\sigma=2$), additive Gaussian white noise (level 2), and downsampling to create a controlled test environment.
  • Real-World Satellite Imagery: The algorithm was tested on actual data from the CBERS-02B (China-Brazil Earth Resources Satellite), providing a practical benchmark for handling real sensor artifacts.

Results and Discussion

The proposed method demonstrates superior visual fidelity and detail preservation compared to traditional models.

Comparison of restoration results: (a) Original Image, (b) Corrupted Image, (c) Adaptive Reciprocal Cell, (d) TV model, and (e) Proposed method.

In simulated tests, the ARC-based model successfully recovered repetitive structures, such as rows of buildings or agricultural fields, that were blurred or aliased in the corrupted input. Unlike the TV model, which often produces blocky or overly smooth artifacts, the proposed method produces results that are visibly clearer and more natural. When applied to real-world CBERS-02B imagery, the method showed a significant improvement in perceived resolution, suggesting robustness for operational satellite data processing.

Conclusion

By treating aliasing as a first-class citizen through frequency-domain masking, this method significantly improves satellite image restoration. Future work will focus on incorporating quantitative metrics like PSNR or SSIM, optimizing dictionary selection specifically for satellite textures, and reducing computational overhead for real-time processing.

Summary updated

Abstract

Recently, an emerging method called image sparse representation has attracted more attentions. The method has been proved to be effective in various image processing applications. It is important to note that few sparse representation methods fail to analyze the aliasing in satellite image restoration. To address the problem, firstly, we employ adaptive reciprocal cell as a image quality estimation tool, which can analyze the satellite image degradation factors including aliasing, blur and noise. Then, with the help of the powerful tool, the estimation about the satellite image quality is introduced into the sparse representation model . Experiment results show that our method can produce good quality restored results.
BibTeX
@article{yanfei2014sparse,
  author = {He Yanfei and Jianwei Zhang and Shunfeng Wang and Yuhui Zheng and Jin Wang and Yunjie Chen},
  title = {{Sparse Representation based Satellite Image Restoration Using Adaptive Reciprocal Cell}},
  journal = {International Journal of Multimedia and Ubiquitous Engineering},
  year = {2014},
  doi = {10.14257/ijmue.2014.9.10.33},
}
Figure