Image Denoising Using Wavelet Transformer
Image Denoising Using Wavelet Transformer
Volume 9, Issue 4, April – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24APR1565
Image Denoising using Wavelet Transformer
1
Kalyani Akhade; 2Sakshi Ghodekar; 3Vaishnavi Kapse; 4Anuja Raykar; 5Sonal Wadhvane
Department of Computer Engineering
Sinhgad College of Technology and Science , Narhe, Pune.
Savitribai Phule Pune University, Pune
Abstract:- As digital imaging becomes increasingly Convolutional Neural Networks (CNNs) and other
important in various fields, the demand for effective sophisticated architectures have showcased exceptional
methods to reduce image noise has risen. This study denoising prowess. Leveraging vast datasets, these models
explores a wide range of techniques for denoising images, learn intricate patterns, enabling them to adaptively filter out
including both traditional and modern methods. It noise and generalize effectively across diverse image
examines classical filters, statistical methods, and datasets.
contemporary machine learning algorithms, explaining
their principles, strengths, and weaknesses. Through a One prevalent deep learning strategy in image
systematic review of existing literature, these techniques denoising involves the utilization of autoencoders and
are categorized based on their underlying approaches variational autoencoders. Trained to encode input images
and practical uses. Comparative analyses offer insights into latent space representations and decode them back to
into the advantages and drawbacks of each method. their original forms while minimizing reconstruction errors,
Additionally, the paper discusses current trends and these models effectively denoise unseen images by learning
future directions in image denoising research. This the underlying structure from clean image data.
comprehensive study serves as a valuable resource for
researchers, professionals, and enthusiasts seeking a Another notable approach employs generative
deep understanding of the evolving field of image adversarial networks (GANs) for image denoising. These
denoising. frameworks consist of a generator and a discriminator
engaged in a competitive learning process. While the
Keywords:- Wavelet Transformer, Image Denoising, generator aims to produce denoised images, the
Machine Learning. discriminator distinguishes between clean and denoised
images. Through this adversarial training, GANs generate
I. INTRODUCTION realistic, high-quality denoised images..
In the realm of digital image processing, the importance Relevance
of image denoising techniques cannot be overstated. Their Image denoising methods are indispensable across a
primary goal is to refine image quality by eliminating broad spectrum of fields, addressing diverse real-world
undesirable noise and artifacts that may arise during image challenges. In the realm of medical imaging, denoising
capture or transmission. Such noise not only compromises elevates diagnostic precision, while in surveillance and
visual clarity but also impacts the precision of subsequent security sectors, it bolsters object recognition and minimizes
image analyses. Employing a variety of algorithms and false alarms. Applications like satellite and remote sensing
strategies, image denoising methods effectively mitigate or benefit from denoising for precise environmental
eliminate noise, thus revealing clearer, more informative monitoring. Within photography, filmmaking, and
images. entertainment, denoising is pivotal for enhancing image
quality, particularly in dimly lit environments. Biomedical
Traditional approaches to image denoising typically imaging relies on denoising for clearer cellular imaging and
rely on mathematical operations, statistical assessments, and advanced medical research. Industries depend on denoising
filter-centric methodologies. These methods are designed to for quality control in industrial inspection, while
detect and suppress noise while safeguarding crucial image autonomous vehicles require pristine visual input for
features. Among them are commonly used filters like accurate decision-making. Digital forensics, document
Gaussian, median, and bilateral filters, each boasting its own imaging, and communication systems all harness denoising
set of strengths and weaknesses. for refined analysis, enhanced OCR accuracy, and improved
image quality during transmission. The significance of
In recent years, the landscape of image denoising has image denoising techniques in these contexts underscores
witnessed a transformative shift fueled by the rise of their indispensable role in delivering precise and dependable
machine learning and deep learning techniques. outcomes across various sectors.
IJISRT24APR1565 www.ijisrt.com 3071
Volume 9, Issue 4, April – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24APR1565
Motivation Presents the SURE-LET approach to image denoising,
The core objective of an image denoising initiative is to which relies on learned experts in the transform domain
mitigate noise across a range of applications, fostering using Stein's unbiased risk estimate. This method
improved visual fidelity and diagnostic precision in medical prioritizes denoising while preserving crucial structural
imaging. It seeks to optimize images for dimly lit scenarios, information in images.
refine data quality for machine learning models, ensure Focuses on image denoising using sparse and redundant
rigorous quality control in industrial workflows, and bolster representations via learned dictionaries. By leveraging
safety measures in autonomous vehicle operations. sparse coding and learned dictionaries, it underscores the
Furthermore, the project endeavors to elevate user significance of sparse representations in capturing
engagement in entertainment, facilitate forensic scrutin y in essential image features during the denoising process,
legal proceedings, and streamline communication systems contributing to the broader field of sparse signal
by minimizing noise in transmitted images. Through these processing.
endeavors, the overarching aim is to tackle critical hurdles, Introduces a sparsity-based approach to image denoising
elevating image quality and data integrity across varied by combining dictionary learning and structural
sectors. clustering. This method synergizes both techniques to
enhance the denoising process effectively.
II. LITERATURE SURVEY Introduces WINNet, an invertible neural network
designed for image denoising, drawing inspiration from
Introduces a fresh method named "unprocessing" aimed wavelet transforms. Incorporating multi-resolution
at enhancing raw image denoising by reversing the initial analysis and sparse representations inspired by wavelet
processing steps applied to raw images. This reversal techniques, it ensures invertibility in the neural network,
allows for more effective denoising using learned potentially advancing image denoising technology with
models, potentially improving performance significantly. applications in various domains such as medical imaging,
Presents an innovative deep raw image denoising remote sensing, and computer vision.
solution tailored specifically for mobile devices, taking
into account their computational limitations. It tackles III. METHODOLOGY
the challenges of deploying deep denoising models on
devices with restricted resources, ensuring efficient A. Data Collection:
performance on mobile platforms. Develop user-friendly interfaces enabling users to
Introduces BM3D-AMP, a pioneering image recovery effortlessly upload noisy images for denoising. Gather pertinent
algorithm that integrates the BM3D denoising technique image details like dimensions, format, and noise attributes.
into the image recovery process. By combining denoising Establish robust data storage mechanisms to securely store
methods with image recovery, it showcases improved uploaded images and metadata in a database. Employ
effectiveness, potentially enhancing overall performance encryption techniques to safeguard sensitive image data during
in image processing tasks. transmission and storage, ensuring confidentiality and
Proposes a method for training a deep convolutional maintaining data integrity.
neural network (CNN) as a denoiser prior for image
restoration, addressing the complexities of learning B. Feature Extraction:
denoiser priors from data. This approach demonstrates Utilize Wavelet Transform for extracting features from
the efficacy of utilizing deep CNNs as denoiser priors in noisy images. Apply discrete wavelet transform (DWT) to
various image restoration tasks. break down images into distinct frequency bands. Utilize
Introduces a novel residual learning framework for deep wavelet thresholding algorithms to eliminate noise and enhance
CNNs to elevate image denoising capabilities beyond image features. Experiment with a variety of wavelet filters and
traditional Gaussian models. By adopting a residual thresholds to optimize denoising performance. Implement
learning approach, the network learns residual denoising methods to address edge effects and preserve image details
functions instead of direct mappings, potentially throughout the denoising process.
enhancing the performance of deep CNNs in denoising
tasks. C. Image Denoising Algorithm:
Addresses the challenge of blind image denoising, where Segment the noisy image into wavelet coefficients using
noise levels are unknown, by proposing a convolutional an appropriate wavelet transform. Utilize techniques to estimate
blind denoising model. This model can effectively handle the underlying image from these coefficients. Reconstruct the
real photographs with varying and unknown noise denoised image from modified wavelet coefficients using
characteristics, extending the applicability of CNN-based inverse wavelet transformation.
denoising methods to real-world scenarios.
IJISRT24APR1565 www.ijisrt.com 3072
Volume 9, Issue 4, April – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24APR1565
D. Implementation and Deployment: E. Evaluation and Validation:
Develop denoising software using Python, with a focus on Validate the denoising method across diverse image
wavelet-based techniques. Utilize efficient algorithms for datasets containing various types and levels of noise. Evaluate
implementing wavelet transforms. Design user-friendly denoising performance using quantitative metrics and visual
interfaces for seamless integration into existing image inspection of denoised images.
processing workflows or applications.
IV. ARCHITECTURE
Fig 1: Block Diagram of Wavelet Transformer
IJISRT24APR1565 www.ijisrt.com 3073
Volume 9, Issue 4, April – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24APR1565
V. RESULTS
Fig 2: Image Denoising App
VI. FUTURE SCOPE REFERENCES
The future prospects for image denoising hold great [1]. T. Brooks, B. Mildenhall, T. Xue, J. Chen, D. Sharlet,
promise, with several key avenues for exploration. This and J. T. Barron, “Unprocessing images for learned raw
includes integrating deep learning methods like CNNs and denoising,” in Proc.IEEE/CVF Conf. Comput. Vis.
GANs to develop more advanced denoising models. Pattern Recognit. (CVPR), Jun. 2019,pp. 11036–
Advancements in end-to-end learning systems and the 11045.
utilization of adversarial training techniques aim to bolster [2]. Y. Wang, H. Huang, Q. Xu, J. Liu, Y. Liu, and J.
the resilience of these models. Future investigations might Wang, “Practical deep raw image denoising on
delve into dynamic noise modeling, capable of adapting to mobile devices,” in Proc. Eur. Conf. Comput.Vis.
various noise patterns in real-time, as well as expanding (ECCV), Aug. 2020, pp. 1–16.
denoising techniques to accommodate multimodal and [3]. C. A. Metzler, A. Maleki, and R. G. Baraniuk,
multispectral imaging data. Additionally, there's a growing “BM3D-AMP: A new image recovery algorithm
interest in real-time and hardware implementations to based on BM3D denoising,” in Proc. IEEE Int. Conf.
support applications such as video streaming and Image Process. (ICIP), Sep. 2015, pp. 3116–3120.
surveillance, along with the development of domain-specific [4]. K. Zhang, W. Zuo, S. Gu, and L. Zhang, “Learning
denoising models tailored to different imaging domains. deep CNN denoiser prior for image restoration,” in
Further exploration into explainable AI, self-supervised Proc. IEEE Conf. Comput. Vis. Pattern Recognit.
learning, and the fusion of traditional and deep learning (CVPR), Jul. 2017, pp. 3929– 3938.
methodologies will play pivotal roles in advancing image [5]. K. Zhang, W. Zuo, Y. Chen, D. Meng, and L. Zhang,
denoising technologies continuously. “Beyond a Gaussian denoiser: Residual learning of
deep CNN for image denoising,”IEEE Trans. Image
VII. CONCLUSION Process., vol. 26, no. 7, pp. 3142–3155, Jul. 2017.
[6]. S. Guo, Z. Yan, K. Zhang, W. Zuo, and L. Zhang,
Thorough analysis of 10 academic papers was “Toward convolutional blind denoising of real
conducted, summarizing notable findings in the literature photographs,” in Proc. IEEE/CVF Conf. Comput.Vis.
review. The study aimed at identifying shortcomings in Pattern Recognit. (CVPR), Jun. 2019, pp. 1712–1722.
current understanding, taking into account the formulation [7]. T. Blu and F. Luisier, “The SURE-LET approach to
of the problem statement and its associated objectives. image denoising,”IEEE Trans. Image Process., vol.
Furthermore, a meticulously crafted action plan was 16, no. 11, pp. 2778–2786, Nov. 2007.
delineated. The developed system is tailored to address the
requirements of end-users comprehensively.
IJISRT24APR1565 www.ijisrt.com 3074
Volume 9, Issue 4, April – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24APR1565
[8]. M. Elad and M. Aharon, “Image denoising via sparse
and redundant representations over learned
dictionaries,” IEEE Trans. Image Process.,vol. 15, no.
12, pp. 3736–3745, Dec. 2006.
[9]. W. Dong, X. Li, L. Zhang, and G. Shi, “Sparsity-
based image denoising via dictionary learning and
structural clustering,” in Proc. CVPR,Jun. 2011, pp.
457–464.
[10]. WINNet: Wavelet-Inspired Invertible Network for
Image Denoising Jun-Jie Huang , Member, IEEE, and
Pier Luigi Dragotti , Fellow, IEEE. IEEE
TRANSACTIONS ON IMAGE PROCESSING,
VOL. 31, 2022.
