Improved Fisher MAP Filter for Despeckling of High-Resolution SAR Images Based on Structural Information Detection

Wei Wei, Zengguo Sun, Zhihua Zhang, Rafał Scherer, Robertas DamaševičiusVytautas 2021 · DOI

Abstract

The Fisher distribution is a robust model for high-resolution (HR) synthetic aperture radar (SAR) images, capturing their characteristic high-peaked and heavy-tailed pixel distributions. This paper proposes an improved Maximum A Posteriori (MAP) filter based on Fisher modeling. By integrating structural information detection—including point target identification, adaptive windowing, and directional sub-window selection—the filter significantly enhances speckle suppression while preserving critical fine details and edges. Experimental results on HR SAR imagery demonstrate superior performance in both homogeneous and complex regions compared to traditional filtering methods.

The Challenge of Speckle in High-Resolution SAR

Synthetic Aperture Radar (SAR) is a vital tool for remote sensing, capable of imaging through clouds and darkness. However, SAR images are inherently affected by speckle, a multiplicative noise caused by the interference of reflected waves. In high-resolution (HR) imagery, particularly in urban or complex environments, the statistical nature of pixels becomes increasingly non-Gaussian, exhibiting "high-peaked and heavy-tailed" distributions. Traditional models like the Rayleigh distribution often fail to capture these extremes, leading to filters that either leave excessive noise or blur essential structural details like building boundaries and point targets.

Fisher Statistical Modeling and Parameter Estimation

To address these limitations, the researchers utilize the Fisher distribution. Derived from the physical multiplicative model of speckle, the Fisher model is both mathematically tractable and highly accurate for HR-SAR data.

Figure 1: Comparison of various statistical distributions against the image histogram on a logarithmic scale, highlighting the Fisher distribution's superior fit for the heavy tail.

As shown in the logarithmic scale analysis, the Fisher distribution tracks the "tail" of the pixel histogram—representing bright outliers and complex textures—far more effectively than empirical models like log-normal or Weibull. To implement the filter, the authors employ "second-kind statistics" based on the Mellin transform, which provides more stable parameter estimation for heavy-tailed distributions than standard moments.

The Improved Structural Information Pipeline

The core innovation of this work is the "Improved" structural detection framework, which moves away from a "one-size-fits-all" filtering approach:

  1. Point Target Detection: The algorithm identifies bright, isolated pixels (e.g., metallic objects) and preserves them immediately to prevent the blurring common in standard filters.
  2. Adaptive Windowing: The processing window size adjusts dynamically; it grows in flat, homogeneous areas to maximize noise suppression and shrinks near complex structures to maintain resolution.
  3. Directional Sub-window Selection: In edge regions, the window is divided into eight directional sub-windows. The filter selects the most homogeneous sub-window for calculation, effectively "filtering around" an edge rather than across it, which prevents edge smearing.

Key Results and Visual Performance

The proposed Improved Fisher MAP filter was evaluated against classical Gamma MAP and standard Fisher MAP filters using quantitative metrics: Equivalent Number of Looks (ENL) for noise suppression and Difference of Coefficient of Variations (DCV) for edge preservation.

  • Noise Suppression: The improved filter achieved an ENL of 122.97, significantly outperforming the Gamma MAP (96.24) and the standard Fisher MAP (52.94).
  • Detail Preservation: It recorded the lowest DCV (0.4491), confirming its success in keeping edges sharp while removing noise.
Figure 4: Visual result of the improved Fisher MAP filter, showing significant speckle reduction in homogeneous areas while maintaining crisp edges and point targets.

Qualitative analysis confirms these metrics. The filter effectively subdues speckle in homogeneous regions (like fields or water) while keeping point targets and building boundaries crisp, as seen in the processed high-resolution samples.

Conclusion

By combining the theoretical strengths of the Fisher distribution with a sophisticated structural detection pipeline, the Improved Fisher MAP filter offers a powerful solution for HR-SAR despeckling. It balances the often-conflicting goals of noise reduction and feature preservation, making it highly suitable for advanced remote sensing applications in complex urban environments.

Summary updated

Abstract

Fisher distribution is a popular model for highresolution (HR) synthetic aperture radar (SAR) images due to its high-peaked and heavy-tailed characteristics as well as its theoretical justification and mathematical tractability. Based on the Fisher modeling of SAR images, the maximum a posteriori (MAP) filter is suggested. In the Fisher model, the parameter of image looks is thought to be fixed to correspond to the formation mechanism of multi-look intensity images, and the other two parameters are accur ately assessed from the SAR image based on second-kind statistics. To improve the Fisher MAP filter especially in the aspect of speckle suppression, the Fisher MAP filter based on recognition of structural information is created using point target detection, the adaptive windowing method, homogeneous region detection, and selection of most homogeneous sub-window. The experiments on despeckling of HR SAR images demonstrate that the improved Fisher MAP filter based on structural information detection can suppress speckle in homogenous and edge regions, and effectively preserve fine details, edges, and point targets.
BibTeX
@inproceedings{wei2021improved,
  author = {Wei Wei and Zengguo Sun and Zhihua Zhang and Rafał Scherer and Robertas Damaševičius},
  title = {{Improved Fisher MAP Filter for Despeckling of High-Resolution SAR Images Based on Structural Information Detection}},
  booktitle = {Vytautas},
  year = {2021},
  doi = {10.3966/160792642021032202016},
}
Figure