Modified model of noise filtering in images based on convulsive neural network with the addition of residual connections and attention module

Authors

DOI:

https://doi.org/10.18372/2073-4751.80.19773

Keywords:

neural networks, CNN, SeConvNet, SAP noise, image enhancement, residual neural network

Abstract

In numerous applied image processing tasks, ranging from technical diagnostics to medical imaging, the accurate restoration of images affected by severe salt-and-pepper (SAP) noise is of critical importance. This type of noise introduces isolated but highly intense white or black pixels, which can significantly distort the visual representation and complicate subsequent analysis.

This study presents a modification of the existing SeConvNet convolutional neural network, specifically designed to enhance SAP noise removal. The primary objective is to improve the processing of images with prominent white noise artifacts, where the original SeConvNet model has demonstrated suboptimal performance. The proposed modification incorporates a novel block that combines residual connections with both channel and spatial attention mechanisms (Channel+Spatial Attention). This integration, implemented through the Convolutional Block Attention Module (CBAM), enables the network to emphasize informative channels and critical spatial regions, thereby enhancing the suppression of high-contrast noise artifacts while preserving fine structural details.

Experimental results indicate a notable improvement in image restoration quality, particularly in datasets where noise-affected regions exhibited a high intensity of bright spots within a 5–8% range. The proposed approach was evaluated on MRI scans, confirming its efficacy in handling complex textures and domain-specific noise artifacts. The significance of these findings lies in the potential integration of the enhanced SeConvNet into medical diagnostic applications and other fields where the precise preservation of fine image structures is paramount.

Despite computational constraints, the introduced modifications have led to improved SAP noise suppression, underscoring the strong potential of this approach for further research and practical implementations.

Author Biography

A.O. Lynovskyi, National Technical University of Ukraine “Igor Sikorsky Kyiv Polytechnic Institute”

In numerous applied image processing tasks, ranging from technical diagnostics to medical imaging, the accurate restoration of images affected by severe salt-and-pepper (SAP) noise is of critical importance. This type of noise introduces isolated but highly intense white or black pixels, which can significantly distort the visual representation and complicate subsequent analysis.

This study presents a modification of the existing SeConvNet convolutional neural network, specifically designed to enhance SAP noise removal. The primary objective is to improve the processing of images with prominent white noise artifacts, where the original SeConvNet model has demonstrated suboptimal performance. The proposed modification incorporates a novel block that combines residual connections with both channel and spatial attention mechanisms (Channel+Spatial Attention). This integration, implemented through the Convolutional Block Attention Module (CBAM), enables the network to emphasize informative channels and critical spatial regions, thereby enhancing the suppression of high-contrast noise artifacts while preserving fine structural details.

Experimental results indicate a notable improvement in image restoration quality, particularly in datasets where noise-affected regions exhibited a high intensity of bright spots within a 5–8% range. The proposed approach was evaluated on MRI scans, confirming its efficacy in handling complex textures and domain-specific noise artifacts. The significance of these findings lies in the potential integration of the enhanced SeConvNet into medical diagnostic applications and other fields where the precise preservation of fine image structures is paramount.

Despite computational constraints, the introduced modifications have led to improved SAP noise suppression, underscoring the strong potential of this approach for further research and practical implementations.

References

He K. et al. Deep Residual Learning for Image Recognition. URL: https://arxiv.org/pdf/1512.03385v1.

Zhang K. et al. Beyond a Gaussian Denoiser: Residual Learning of Deep CNN for Image Denoising. URL: https://ieeexplore.ieee.org/document/7839189.

Lu T. et al. Autoencoder Combined with CBAM Improves Denoising of MR Images. URL: https://ieeexplore.ieee.org/document/9750433.

Woo S. et al. CBAM: Convolutional Block Attention Module. URL: https://arxiv.org/pdf/1807.06521v2.

Cai G. CBAM-DnCNN: An Improved Method For Image Denoising. URL: https://www.computer.org/csdl/proceedings-article/eiecs/2023/10435545/1UIC6jLXz0Y.

Rafiee A. A., Farhang M. A deep convolutional neural network for salt-and-pepper noise removal using selective convolutional blocks. URL: https://arxiv.org/pdf/2302.05435v1.

Wang T., Hu Z., Guan Y. An efficient lightweight network for image denoising using progressive residual and convolutional attention feature fusion. URL: https://www.nature.com/articles/s41598-024-60139-x.

Линовський А. О., Мухін В. Є. Модифікована модель фільтрації шуму на зображеннях на основі згорткової нейронної мережі. Кібербезпека: освіта, наука, техніка. 2024. № 4(24). С. 388–397. URL: https://csecurity.kubg.edu.ua/index.php/journal/article/view/634.

Downloads

Published

2025-03-13

Issue

Vol. 4 No. 80 (2024)

Section

Статті

License

Автори, які публікуються у цьому журналі, погоджуються з наступними умовами:

  1. Автори залишають за собою право на авторство своєї роботи та передають журналу право першої публікації цієї роботи на умовах ліцензії Creative Commons Attribution License, котра дозволяє іншим особам вільно розповсюджувати опубліковану роботу з обов'язковим посиланням на авторів оригінальної роботи та першу публікацію роботи у цьому журналі.

  2. Автори мають право укладати самостійні додаткові угоди щодо неексклюзивного розповсюдження роботи у тому вигляді, в якому вона була опублікована цим журналом (наприклад, розміщувати роботу в електронному сховищі установи або публікувати у складі монографії), за умови збереження посилання на першу публікацію роботи у цьому журналі.

  3. Політика журналу дозволяє і заохочує розміщення авторами в мережі Інтернет (наприклад, у сховищах установ або на особистих веб-сайтах) рукопису роботи, як до подання цього рукопису до редакції, так і під час його редакційного опрацювання, оскільки це сприяє виникненню продуктивної наукової дискусії та позитивно позначається на оперативності та динаміці цитування опублікованої роботи (див. The Effect of Open Access).