Project_proposal

Exploring the Effects of Over-Denoising and Under-Denoising in
Image Processing
Hamza Ali
October 10, 2024
1 Introduction
Image denoising is a key task in digital image processing, where noise is reduced to improve image
quality. Noise, introduced by sensor imperfections or transmission errors, degrades visual clarity. Over-
denoising risks removing crucial details, while under-denoising leaves unwanted noise. This project
examines the effects of both, on denoised images.
1.1 Research problem

The problem addressed in this research is the impact of incorrectly estimating noise levels during
denoising. When noise levels are overestimated, it leads to over-denoising, removing significant image
features. Conversely, underestimating noise levels results in under-denoising, retaining undesirable
artifacts. The study will analyze both extremes’ effects on image quality to better understand how
noise misestimations affect final outcomes.
1.2 Background, Motivation, and Purpose

Denoising is used across many fields, such as medical imaging and photography, to enhance image
clarity. Misestimating the noise level can lead to over- or under-denoising, impacting the quality of the
output image. The purpose of this study is to evaluate these effects using state-of-the-art denoising
methods and datasets.
2 Related Work
Deep learning-based denoising methods, such as DRUNET [1], have achieved impressive results under
controlled conditions, where noise levels are known. Studies such as those by Bevilacqua et al. [2] and
Martin et al. [3] have provided datasets and benchmarks for evaluating denoising algorithms. This
study contributes to the field by investigating the consequences of misestimating noise levels during
denoising.
2.1 Prior Art

DRUNET [1] and similar algorithms have demonstrated effective noise reduction when noise levels are
accurately estimated. However, these models often fail when noise is inaccurately estimated, leading
to suboptimal denoising. This research focuses on addressing this gap by systematically evaluating the
effects of over- and under-denoising using publicly available datasets.
1
2.2 Theoretical Framework
The study is based on principle of advanced image processing techniques. Noise is added to the images,
and the effects of denoising are evaluated using Peak Signal-to-Noise Ratio (PSNR) and Structural
Similarity Index (SSIM), two widely used metrics for assessing image quality.
2.3 Placement of Study

The study is positioned within the field of image restoration, extending the current research by exam-
ining how different levels of noise affect the visual quality of images after denoising.
3 Research Question and Hypothesis

How do over- and under-denoising affect denoised image quality, in terms of image quality metrics?
So, over-denoising reduces essential image features, while under-denoising leaves noise artifacts on it.
We aim to explore how these effects influence the overall image quality.
3.1 Expected Results

Over-denoised images are expected to show loss of important details, while under-denoised images will
likely retain too much noise. The comparison between the values of noised and denoised image quality
matrices, will highlight the effects of over- and under-denoising on image quality.
4 Methodology
This research uses publicly available datasets like Set5 and BSDS100 to create noisy images. The
DRUNET algorithm is applied to these images, followed by an evaluation using PSNR and SSIM. The
results will show how over- and under-denoising impact image quality.
4.1 Tools and Algorithms

Artificial noise will be added to images and DRUNET algorithm will be implemented in Python to
denoise them. Standard image formats (JPEG, PNG) will be used for processing, and image quality
will be evaluated using PSNR and SSIM.
4.2 Evaluation Criteria

The images will be evaluated based on PSNR and SSIM scores, as well as visual inspection to assess
noise reduction and detail preservation.
4.3 Analysis Method

Results will be plotted and compared by using PSNR and SSIM values at different noise levels. A
quantitative analysis will be conducted based on these results.
5 Time Plan
• Weeks 40-42: Literature review and dataset preparation.
• Weeks 43-45: Implement DRUNET and noise generation
• Weeks 46-49: Run experiments, gather data and Analyze results
• Weeks 50-51: Submit proposal for review
• Weeks 52-02: Finalize the project and prepare a presentation
• Weeks 02-03: Present and submit project
2
References
[1] K. Zhang, Y. Li, W. Zuo, L. Zhang, L. Van Gool, and R. Timofte, “Plug-and-play image restora-
tion with deep denoiser prior,” IEEE Transactions on Pattern Analysis and Machine Intelligence,
vol. 44, no. 10, pp. 6360–6376, 2022.
[2] M. Bevilacqua, A. Roumy, C. Guillemot, and M. L. Alberi-Morel, Low-complexity single-image
super-resolution based on nonnegative neighbor embedding. BMVA Press, 2012.
[3] D. Martin, C. Fowlkes, D. Tal, and J. Malik, “A database of human segmented natural images
and its application to evaluating segmentation algorithms and measuring ecological statistics,”
in Proceedings Eighth IEEE International Conference on Computer Vision. ICCV 2001, vol. 2,
pp. 416–423 vol.2, 2001.

Project_proposal

Uploaded by

Copyright:

Available Formats

Project_proposal

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Project_proposal

Uploaded by

Copyright:

Available Formats

Exploring the Effects of Over-Denoising and Under-Denoising in

1.1 Research problem

1.2 Background, Motivation, and Purpose

2.1 Prior Art

2.3 Placement of Study

3 Research Question and Hypothesis

3.1 Expected Results

4.1 Tools and Algorithms

4.2 Evaluation Criteria

4.3 Analysis Method

You might also like