Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Evaluating BM3D and NBNet: A Comprehensive Study of Image Denoising Across Multiple Datasets

Ghazal Kaviani, Reza Marzban, Ghassan AlRegib

TL;DR

This paper addresses image denoising across diverse real-world noise conditions by comparing BM3D, a traditional non-learning method, with NBNet, a learning-based approach. It conducts a comprehensive cross-dataset evaluation (CURE-OR, CURE-TSR, SSID+, Set-12, Chest-Xray) using seven IQA metrics and a detection-based downstream task to assess practical impact. NBNet generally excels in complex, non-independent noise (e.g., underexposure/overexposure) while BM3D performs well on blur and certain structured noise; on the SIDD dataset NBNet shows notable gains in $PSNR$ and $SSIM$, whereas Chest-Xray results often favor BM3D due to noise independence. The study demonstrates the importance of dataset-specific denoising choices and multi-metric, task-oriented evaluation for real-world deployment.

Abstract

This paper investigates image denoising, comparing traditional non-learning-based techniques, represented by Block-Matching 3D (BM3D), with modern learning-based methods, exemplified by NBNet. We assess these approaches across diverse datasets, including CURE-OR, CURE-TSR, SSID+, Set-12, and Chest-Xray, each presenting unique noise challenges. Our analysis employs seven Image Quality Assessment (IQA) metrics and examines the impact on object detection performance. We find that while BM3D excels in scenarios like blur challenges, NBNet is more effective in complex noise environments such as under-exposure and over-exposure. The study reveals the strengths and limitations of each method, providing insights into the effectiveness of different denoising strategies in varied real-world applications.

Evaluating BM3D and NBNet: A Comprehensive Study of Image Denoising Across Multiple Datasets

TL;DR

This paper addresses image denoising across diverse real-world noise conditions by comparing BM3D, a traditional non-learning method, with NBNet, a learning-based approach. It conducts a comprehensive cross-dataset evaluation (CURE-OR, CURE-TSR, SSID+, Set-12, Chest-Xray) using seven IQA metrics and a detection-based downstream task to assess practical impact. NBNet generally excels in complex, non-independent noise (e.g., underexposure/overexposure) while BM3D performs well on blur and certain structured noise; on the SIDD dataset NBNet shows notable gains in and , whereas Chest-Xray results often favor BM3D due to noise independence. The study demonstrates the importance of dataset-specific denoising choices and multi-metric, task-oriented evaluation for real-world deployment.

Abstract

This paper investigates image denoising, comparing traditional non-learning-based techniques, represented by Block-Matching 3D (BM3D), with modern learning-based methods, exemplified by NBNet. We assess these approaches across diverse datasets, including CURE-OR, CURE-TSR, SSID+, Set-12, and Chest-Xray, each presenting unique noise challenges. Our analysis employs seven Image Quality Assessment (IQA) metrics and examines the impact on object detection performance. We find that while BM3D excels in scenarios like blur challenges, NBNet is more effective in complex noise environments such as under-exposure and over-exposure. The study reveals the strengths and limitations of each method, providing insights into the effectiveness of different denoising strategies in varied real-world applications.
Paper Structure (4 sections, 5 figures, 12 tables)

This paper contains 4 sections, 5 figures, 12 tables.

Table of Contents

  1. Introduction
  2. Datasets
  3. Methodology
  4. Discussion

Figures (5)

  • Figure 1: Sample images from the Chest X-ray dataset depicting a case of pneumonia. The four panels include (a) the noisy input image, (b) the original image, (c) the image denoised using NBNet, and (d) the image denoised using BM3D.
  • Figure 2: NBNet employs subspace projection for denoising, generating basis for the signal subspace. This figure illustrates its effectiveness on level 3 noise from the CURE-OR dataset.
  • Figure 3: Sample image denoised by NBNet model
  • Figure 4: Correlation Matrix of IQA Metrics. The heatmap illustrates the correlation between different Image Quality Assessment (IQA) metrics. Each cell displays the correlation coefficient, with warmer colors indicating stronger correlations. A larger font size is used for better readability, and the color bar numbers are also enlarged for clarity.
  • Figure 5: SHAP Summary Bee Swarm and Bar Plot showing feature distribution and importance in the model.