Reinforced Swin-Convs Transformer for Underwater Image Enhancement

Tingdi Ren,Haiyong Xu,Gangyi Jiang,Mei Yu,Ting Luo

DOI: https://doi.org/10.48550/arXiv.2205.00434

2022-05-01

Abstract:Underwater Image Enhancement (UIE) technology aims to tackle the challenge of restoring the degraded underwater images due to light absorption and scattering. To address problems, a novel U-Net based Reinforced Swin-Convs Transformer for the Underwater Image Enhancement method (URSCT-UIE) is proposed. Specifically, with the deficiency of U-Net based on pure convolutions, we embedded the Swin Transformer into U-Net for improving the ability to capture the global dependency. Then, given the inadequacy of the Swin Transformer capturing the local attention, the reintroduction of convolutions may capture more local attention. Thus, we provide an ingenious manner for the fusion of convolutions and the core attention mechanism to build a Reinforced Swin-Convs Transformer Block (RSCTB) for capturing more local attention, which is reinforced in the channel and the spatial attention of the Swin Transformer. Finally, the experimental results on available datasets demonstrate that the proposed URSCT-UIE achieves state-of-the-art performance compared with other methods in terms of both subjective and objective evaluations. The code will be released on GitHub after acceptance.

Computer Vision and Pattern Recognition,Image and Video Processing

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the challenges in underwater image enhancement (UIE), namely, restoring the quality of underwater images that has deteriorated due to light absorption and scattering. Specifically, the paper proposes a U - Net - based enhanced Swin - Convs Transformer method (URSCT - UIE) aiming to improve the quality of underwater images. Underwater imaging is affected by particles in water, resulting in a decline in image quality during light propagation, such as color deviation, color distortion and blurry details. These problems are also common in images taken by high - end cameras, so effective technical means are required to enhance the quality of underwater images.

The main contributions of the paper are as follows:
1. **Improving global dependency capture**: By embedding the Swin Transformer into U - Net, the ability to capture global dependencies is enhanced, which is an indispensable part of the UIE task.
2. **Fusing convolution and core attention mechanism**: Since the pure Swin Transformer is weak in capturing local attention, the paper proposes an ingenious method to fuse convolution with the core attention mechanism of the Swin Transformer and constructs an enhanced Swin - Convs Transformer Block (RSCTB) to strengthen local attention.
3. **Achieving state - of - the - art performance**: The proposed URSCT - UIE method has achieved state - of - the - art performance through full - reference quantitative metrics in multiple recent benchmark tests.

Through these improvements, the method proposed in the paper is not only innovative in technology but also shows excellent performance in practical applications, especially when dealing with complex and dynamic underwater environments.