Using Anisotropic Diffusion for Coherence Estimation

François Aspert, Alessio Cantone, Jean‐Philippe Thiran, Paolo PasqualiInfoscience 2007

TL;DR

This paper introduces novel methods for estimating coherence in SAR interferometry data, crucial for accurate geological and land-use analysis. By employing anisotropic diffusion, the authors significantly reduce bias in low-coherence regions while preserving fine spatial details, outperforming traditional boxcar methods.

What Problem Does This Paper Address?

Synthetic Aperture Radar (SAR) interferometry (InSAR) relies heavily on coherence estimation to assess the quality of interferograms. This quality measure is vital for applications like precise displacement and elevation mapping, and land-use classification. However, existing coherence estimators suffer from a bias towards higher values, especially in areas with low coherence. While increasing the number of samples used in estimation can mitigate this bias, it comes at the cost of reduced spatial resolution. This is particularly problematic for medium-resolution sensors like ASAR WS, where target variability is high. The paper tackles this challenge by developing a coherence estimation technique that adapts its processing window to local image characteristics, aiming to remove bias without sacrificing spatial detail.

Methods

The core of the paper's contribution lies in the application of anisotropic diffusion for coherence estimation.

  • Anisotropic Diffusion: This is a powerful image filtering technique that smooths images while preserving important features like edges. Unlike traditional diffusion that smooths uniformly, anisotropic diffusion adapts its smoothing based on local image gradients. The paper extends this concept to vector-valued images, allowing it to process multiple SAR images simultaneously.
  • Amplitude Driven Diffusion (ADD): This method uses the intensity (amplitude) of the SAR images to guide the diffusion process. By analyzing the gradients in the amplitude images, it identifies areas of discontinuity and uses this information to adapt the diffusion coefficients. This helps in identifying homogenous regions for reliable coherence estimation.
  • Boxcar Driven Diffusion (BDD): This approach uses a preliminary, low-resolution coherence map (derived from a simple boxcar estimation) to guide the anisotropic diffusion. This allows the diffusion process to be sensitive to variations in coherence, even if the amplitude images appear similar.
  • Combination Method: The authors propose a novel combination of ADD and BDD. This hybrid approach leverages the strengths of both methods: ADD handles cases where amplitude and coherence behave oppositely, while BDD excels at discriminating regions with similar backscattering coefficients but different coherence. The final coherence is determined by taking the maximum of the ADD and BDD estimates, with a threshold to ensure robustness.

Datasets

The paper evaluates its methods on two datasets:

  • ERS amplitude images and their corresponding phase-compensated interferogram: This dataset originates from the Bienne region in Switzerland, covering lakes and coastal zones.
  • ENVISAT-WS power images and their corresponding flattened interferogram: This dataset is also used for evaluation.

The paper does not introduce new benchmark datasets but rather uses existing SAR interferometry data to demonstrate the effectiveness of its proposed techniques.

Key Insights and Ideas

  • Anisotropic diffusion is a powerful tool for coherence estimation: By adapting smoothing based on local image geometry, it can overcome the resolution-vs-bias trade-off inherent in traditional methods.
  • Leveraging amplitude and coherence information separately and combined is key: ADD effectively handles amplitude variations, while BDD is sensitive to coherence changes. Combining them provides a more robust and comprehensive estimation.
  • The bias in low-coherence regions is a significant problem: Traditional methods struggle here, and the proposed anisotropic diffusion approaches offer a substantial improvement.
  • Preserving spatial resolution is achievable: The anisotropic nature of the diffusion allows for sharp transitions and detailed feature preservation, which is critical for accurate InSAR applications.

Key Results

  • The proposed methods (ADD and BDD) significantly reduce the bias in low-coherence regions compared to standard 3x3 boxcar estimation.
  • The combined ADD-BDD method demonstrates improved overall coherence map reliability by effectively handling cases where amplitude and coherence behave oppositely.
  • Qualitative results show that the anisotropic diffusion methods preserve fine spatial details, such as river channels, while smoothing out noise in homogenous areas like lakes.
  • The paper presents comparative coherence profiles and histograms, visually demonstrating the improved smoothing and bias reduction achieved by the proposed techniques.

Limitations and Future Work

  • Parameter tuning: While the paper mentions using a median-based threshold and an anisotropy parameter, the optimal selection of these parameters for various scenarios might require further investigation.
  • Complex cases: The authors acknowledge that their combined method is a "preliminary attempt to solve complex cases" and that "some research is needed to find an optimal was to combine both techniques."
  • Further exploration: The paper suggests that while promising results have been achieved, "several ways can still be explored to improve the results accuracy." This implies potential for further refinement of the diffusion models, combination strategies, and parameter optimization.

Summary updated

Abstract

In this paper we will present a new coherence estimation technique for SAR interferometry products that adapts the estimation window size and shape during processing. This is of particular interest for sensors with medium spatial resolution, like the ASAR WS mode, where the estimator shall cope with the spatial variability of the targets in the imaged area. This method is designed to remove low coherence magnitude bias while keeping a good spatial resolution. Finally, this new approach will be compared to an existing algorithm for quantifying resulting quality improvement. 1.
BibTeX
@inproceedings{aspert2007using,
  author = {François Aspert and Alessio Cantone and Jean‐Philippe Thiran and Paolo Pasquali},
  title = {{Using Anisotropic Diffusion for Coherence Estimation}},
  booktitle = {Infoscience},
  year = {2007},
}
Figure