Speckle Noise Reduction in Sonar Image Based on Adaptive Redundant Dictionary

Yifan Huang, Weixiang Li, Fei YuanJournal of Marine Science and Engineering 2020 · DOI

TL;DR

This paper introduces a novel adaptive dictionary learning method that combines K-SVD with wavelet transforms to effectively reduce speckle noise in sonar images. The proposed approach offers superior speckle noise removal and edge detail preservation compared to existing methods, while significantly improving computational efficiency, making it well-suited for onboard AUV processing.

What Problem Does This Paper Address?

Sonar images are crucial for underwater exploration and operations, but they are often degraded by speckle noise. This noise arises from the complex scattering and reverberation of acoustic waves in water, particularly from seafloor sediments. While deep learning methods have shown promise in image denoising, their high computational demands and large data requirements make them impractical for resource-constrained Autonomous Underwater Vehicles (AUVs) that collect sonar data. Traditional spatial and transform domain filtering methods can also lead to loss of image details. Therefore, there's a need for efficient and effective denoising techniques that can be deployed on AUVs.

Methods

The authors propose an adaptive dictionary learning method that leverages both K-SVD and wavelet transforms. The core idea is to create a dictionary that captures multi-resolution characteristics of the sonar image. This is achieved by:

  • Wavelet Transform: Applying a multi-layer wavelet transform to decompose the noisy sonar image into different frequency bands (low-frequency and high-frequency coefficients at various scales).
  • K-SVD Dictionary Learning: Using the extracted wavelet coefficients to train an adaptive dictionary. K-SVD is employed for its ability to iteratively update dictionary atoms and sparse representation coefficients, effectively learning a representation of the image while suppressing noise.
  • Multi-resolution Dictionary: The resulting dictionary contains atoms at different resolution scales. Large-scale atoms capture the overall image structure, while small-scale atoms preserve fine details. This multi-resolution dictionary is then used for denoising.
  • Logarithmic Transformation: To handle the multiplicative nature of speckle noise in sonar images, a logarithmic transformation is applied, converting it into an approximately additive Gaussian noise problem, which is then addressed by the dictionary learning approach. An inverse logarithmic transform is applied to recover the denoised image.

Datasets

The paper uses two types of datasets for evaluation:

  • Simulated Sonar Images: Speckle noise is synthetically added to clean sonar images using a method described in Section 2, which simulates speckle noise characteristics well.
  • Real Sonar Images: Images are sourced from the Fenn Enterprises side scan sonar image library. These images are grayscaled and resized to 256x256 for processing.

Key Insights and Ideas

  • Bridging the Gap: The paper successfully bridges the gap between computationally intensive deep learning methods and less effective traditional filters by proposing a dictionary learning approach that is both efficient and effective for sonar image denoising.
  • Multi-resolution Advantage: Incorporating wavelet transform into dictionary learning allows the dictionary to capture image information at multiple scales, leading to better preservation of both overall structure and fine details.
  • AUV Suitability: The proposed method's reduced computational complexity makes it a strong candidate for onboard processing in AUVs, enabling real-time denoising without requiring massive computational resources.
  • Speckle Noise Modeling: The paper provides a clear explanation of the statistical model of sonar speckle noise, particularly its multiplicative nature, and how it can be transformed for processing.

Key Results

  • Superior Denoising: The proposed method consistently outperforms classical methods (Lee, Kuan, Frost, SRAD, Wavelet, Curvelet, DCT) and the standard K-SVD algorithm in terms of both speckle noise reduction and edge detail preservation, as evidenced by objective metrics (SSIM, ENL) and visual inspection of denoised images.
  • Significant Speedup: The proposed algorithm is approximately twice as fast as the K-SVD algorithm. For example, at a noise standard deviation of 0.9, the proposed method took 146.81 seconds, while K-SVD took 315.76 seconds. This substantial reduction in computation time is a key advantage for AUV applications.
  • Objective Evaluation: Tables 1-3 and Figure 6 demonstrate that the proposed method achieves the highest SSIM values (indicating better image quality and structure preservation) and ENL values (indicating better speckle suppression) across various noise levels.

Limitations and Future Work

  • Limitations: The paper acknowledges that while the proposed method is efficient, it still relies on dictionary learning, which can be more complex than simpler filtering techniques. The performance is also dependent on the quality of the learned dictionary.
  • Future Work: The authors suggest that the proposed method effectively reduces computation time and improves efficiency. Further research could explore more advanced dictionary learning techniques or hybrid approaches to further enhance denoising performance and robustness in even more challenging underwater environments. The paper also implicitly suggests that further optimization of the wavelet decomposition layers and K-SVD parameters could yield even better results.

Summary updated

Abstract

As acoustic waves are affected by the channel characteristics, such as scattering and reverberation when propagating in water, sonar images often exhibit speckle noise which will cause visual quality of the image to decrease. Therefore, denoising is a crucial preprocessing technique in sonar image applications. However, speckle noise is mainly caused by the sediment echo signals which are related to the background of seafloor sediment and can be obtained by prior modeling. Although deep learning-based denoising algorithms represent a research hotspot now, they are not suitable for such applications due to the high calculation amount and the large requirement of original images considering that sonar is carried by Autonomous Underwater Vehicles (AUVs) for collecting sonar images and performing calculation. In contrast, dictionary learning-based denoising method is more suitable and easier to be modeled. Compared with deep learning, it can greatly reduce the calculation amount and is more easily integrated into AUV systems. In addition, dictionary learning method based on image sparse representation can effectively achieve image denoising similarly. In order to solve the above problems, we propose a new adaptive dictionary learning method based on multi-resolution characteristics, which combines K-SVD dictionary learning with wavelet transform. Our method has the characteristics of dictionary learning and inherits the features of wavelet analysis as well. Compared with several classical methods, the proposed method is better at speckle noise reduction and edge detail preservation. At the same time, the calculation time is greatly reduced and the efficiency is significantly improved.
BibTeX
@article{huang2020speckle,
  author = {Yifan Huang and Weixiang Li and Fei Yuan},
  title = {{Speckle Noise Reduction in Sonar Image Based on Adaptive Redundant Dictionary}},
  journal = {Journal of Marine Science and Engineering},
  year = {2020},
  doi = {10.3390/jmse8100761},
}
Figure