A Denoising Method for Randomly Clustered Noise in ICCD Sensing Images Based on Hypergraph Cut and Down Sampling

Meng Yang, Fei Wang, Yibin Wang, Nanning Zheng

TL;DR

This paper introduces a novel denoising method specifically designed for Intensified Charge-Coupled Device (ICCD) images, which suffer from a unique type of spatially clustered noise that traditional filters struggle with. By leveraging down-sampling and hypergraph cuts to identify and isolate noisy patches, the method effectively removes this challenging noise while preserving crucial image details.

What Problem Does This Paper Address?

Intensified Charge-Coupled Device (ICCD) sensors are crucial for capturing images in extremely low-light conditions. However, the image intensification process introduces a specific type of noise: spatially clustered noise. This noise is problematic because it doesn't conform to the assumptions of standard denoising algorithms, which typically handle independent and identically distributed (i.i.d.) noise or fixed patterns. The clustered noise in ICCD images can be random in its appearance and scale, often masking or distorting the true structural information within the image. Existing denoising methods either fail to remove this noise effectively or, worse, can even amplify it by mistaking it for genuine image features. This paper aims to bridge this gap by developing a method that can accurately identify and remove this specific type of clustered noise while preserving image quality.

Methods

The proposed denoising algorithm consists of several key steps:

  1. Over-segmentation: The ICCD image is first segmented into small, non-overlapping "flat patches" using the watershed algorithm. The goal is to create patches that ideally contain minimal structural information, making them easier to analyze for noise.
  2. Residual Estimation via Down-sampling and Up-sampling: A core insight of this paper is that the scale of the clustered noise is often smaller than that of the true structural information. To exploit this, the image is down-sampled and then up-sampled. By comparing this processed image with the original noisy image, the "residuals" are calculated. These residuals highlight areas where the noise is most prominent. The down-sampling scale is adaptively determined for each patch based on its estimated scale, rather than using a fixed value.
  3. Hypergraph Construction and Cut: The patches are then treated as vertices in a hypergraph. Hyperedges are constructed to represent relationships between patches, incorporating information from the estimated residuals, patch positions, and gray-level values. A hypergraph cut algorithm is then employed to classify these patches into two categories: "noisy patches" and "noise-free patches." This classification is crucial for applying different processing strategies to each type of patch.
  4. Patch-wise Filtering:
    • Noise-free patches: These patches are processed using a standard, effective denoising algorithm like Block-Matching and 3D filtering (BM3D), as they are assumed to contain little to no clustered noise.
  5. Noisy patches: The identified clustered noise is estimated by subtracting the mean residual value from the noisy patch. The recovered noisy patch is then obtained by subtracting this estimated noise.
  6. Post-processing with RPCA: After the initial patch-wise filtering, some sparse clustered noise might remain. To address this, Robust Principal Component Analysis (RPCA) is applied as a post-processing step. RPCA helps to decompose the image into a low-rank component (representing structure) and a sparse component (representing noise), further refining the denoised image. The parameter for RPCA is adaptively adjusted based on the image content to balance noise removal and detail preservation.

Datasets

The paper evaluates its method on a dataset of real ICCD images. The authors mention that the ground truth for these images is not available, so the evaluation is primarily subjective. They also refer to experiments with simulated noise on ICCD images for parameter tuning.

Key Insights and Ideas

  • Scale Difference is Key: The fundamental insight is that the clustered noise in ICCD images is often significantly smaller in scale than the actual image structures. This observation is the basis for the down-sampling/up-sampling residual estimation technique.
  • Hypergraphs for Complex Relationships: Traditional graph-based methods struggle with complex relationships between image patches. The use of hypergraphs allows for a more nuanced modeling of these relationships, incorporating multiple attributes (residuals, position, intensity) to improve patch classification.
  • Adaptive RPCA for Balanced Denoising: Applying RPCA with a fixed parameter can lead to either over-smoothing or insufficient noise removal. The paper's adaptive RPCA, which adjusts a key parameter based on image features, strikes a better balance between preserving details and eliminating residual noise.
  • Patch-wise Processing for Targeted Noise Removal: By classifying patches and treating noisy and noise-free patches differently, the algorithm can apply specialized techniques, leading to more effective and targeted noise reduction.

Key Results

  • The proposed algorithm effectively removes randomly clustered noise from ICCD images.
  • It preserves true textural information better than existing methods, including the algorithm from reference [2] and general denoising algorithms like BM3D and BLS-GSM.
  • General denoising methods (BM3D, BLS-GSM) can sometimes amplify the clustered noise in ICCD images.
  • The K-SVD method can remove some noise but may introduce significant structural distortion.
  • The proposed algorithm, even with RPCA, is significantly faster than the algorithm in [2] (saving over 30% runtime). Without RPCA, it's about 10 times faster.

Limitations and Future Work

  • Ground Truth Availability: The primary limitation noted is the lack of ground truth for real ICCD images, necessitating subjective evaluation.
  • Parameter Tuning: While the hypergraph method is shown to be robust, determining optimal parameters (like n, m, and α) still requires experimental tuning.
  • Computational Cost: While faster than some alternatives, the RPCA step still contributes significantly to the overall runtime. Future work could focus on further optimizing this step or exploring alternative post-processing methods.
  • Generalizability: The method is specifically tailored for ICCD images and their particular noise characteristics. Its effectiveness on other types of images with different noise profiles would need to be investigated.

Summary updated

Abstract

Intensified charge-coupled device (ICCD) images are captured by ICCD sensors in extremely low-light conditions. They often contains spatially clustered noises and general filtering methods do not work well. We find that the scale of the clustered noise in ICCD sensing images is often much smaller than that of the true structural information. Then the clustered noise can be identified by properly down-sampling and then up-sampling the ICCD sensing image and comparing it to the noisy image. Based on this finding, we present a denoising algorithm to remove the randomly clustered noise in ICCD images. First, we over-segment the ICCD image into a set of flat patches, and each patch contains very little structural information. Second, we classify the patches into noisy patches and noise-free patches based on the hypergraph cut method. Then the noise-free patches are easily recovered by the general block-matching and 3D filtering (BM3D) algorithm, since they often do not contain the clustered noise. The noisy patches are recovered by subtracting the identified clustered noise from the noisy patches. After that, we could get the whole recovered ICCD image. Finally, the quality of the recovered ICCD image is further improved by diminishing the remaining sparse noise with robust principal component analysis. Experiments are conducted on a set of ICCD images and compared with four existing denoising algorithms, which shows that the proposed algorithm removes well the randomly clustered noise and preserves the true textural information in the ICCD sensing images.
BibTeX
@inproceedings{yang2017denoising,
  author = {Meng Yang and Fei Wang and Yibin Wang and Nanning Zheng},
  title = {{A Denoising Method for Randomly Clustered Noise in ICCD Sensing Images Based on Hypergraph Cut and Down Sampling}},
  booktitle = {Sensors},
  year = {2017},
  url = {https://openreview.net/forum?id=PnzrNrqIYG},
}
Figure