Multiscale Attention Fusion for Underwater Image Enhancement
Barkat, Haris Ali (2024)
2024
All rights reserved. This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:amk-2024112931288
https://urn.fi/URN:NBN:fi:amk-2024112931288
Tiivistelmä
In recent years, underwater exploration has made underwater image processing an evolving research domain. Ocean resource exploration, marine ecology, underwater microscopic detection, deep-sea installation monitoring, terrain scanning and detection of mines and telecommunication cables, self-propelled underwater vehicles, and naval military applications all rely on underwater imaging analysis. However, problems are associated with underwater images due to the haze, color, and contrast distributed unevenly in different regions within a given image. That requires advanced image enhancement techniques to get the most information from images.
To overcome these problems and challenges, a multiscale attention fusion for underwater image enhancement is proposed. This model utilizes several key modules, including the Residual Dual Feature Attention Block (RDFAB), the Selective Channels Kernel Feature Fusion Block (SCKFFB), the Multiscale Residual Block (MRB), and the Restoration Block (RB). RDFAB utilizes residual block, channel attention, and spatial attention. The residual block learns residual information from the input image and focuses on subtle details from the noisy image. Channel Attention Block captures the inter-dependence across different color channels in a feature map. The Spatial Attention Block aids in understanding spatial relations among image pixels and enables the model to pay more attention to the spatial regions. SCKFFB helps merge features in different color channels efficiently. MRB also contributes to capturing multiscale features from the input image. Finally, by exploiting the features extracted and processed by the above-mentioned modules, the restoration block leverages the learned information and helps refine these features for final image quality and clarity. The experiment results show that the proposed model has significantly improved the underwater image quality compared with the state-of-the-art prior and deep learning-based methods by retaining the color contrast, noise reduction, and minimizing the artifacts in the image.
To overcome these problems and challenges, a multiscale attention fusion for underwater image enhancement is proposed. This model utilizes several key modules, including the Residual Dual Feature Attention Block (RDFAB), the Selective Channels Kernel Feature Fusion Block (SCKFFB), the Multiscale Residual Block (MRB), and the Restoration Block (RB). RDFAB utilizes residual block, channel attention, and spatial attention. The residual block learns residual information from the input image and focuses on subtle details from the noisy image. Channel Attention Block captures the inter-dependence across different color channels in a feature map. The Spatial Attention Block aids in understanding spatial relations among image pixels and enables the model to pay more attention to the spatial regions. SCKFFB helps merge features in different color channels efficiently. MRB also contributes to capturing multiscale features from the input image. Finally, by exploiting the features extracted and processed by the above-mentioned modules, the restoration block leverages the learned information and helps refine these features for final image quality and clarity. The experiment results show that the proposed model has significantly improved the underwater image quality compared with the state-of-the-art prior and deep learning-based methods by retaining the color contrast, noise reduction, and minimizing the artifacts in the image.