Underwater Image Enhancement via Adaptive Group Attention-Based Multiscale Cascade Transformer

Zhixiong Huang,Jinjiang Li,Zhen Hua,Linwei Fan
DOI: https://doi.org/10.1109/tim.2022.3189630
IF: 5.6
2022-07-22
IEEE Transactions on Instrumentation and Measurement
Abstract:The absorption and scattering caused by the underwater medium degrade the quality of underwater optical imaging, which limits the further development of underwater tasks. Recently, transformer-based methods have shown the same excellent performance as convolutional neural networks (CNNs) in various vision tasks, but the huge parameters of such networks hinder their application deployment. In this article, we propose novel adaptive group attention (AGA), which can dynamically select visually complementary channels based on the dependencies, reducing the number of further attention parameters. The AGA is applied in the Swin Transformer module and used to design an end-to-end underwater image enhancement network. The network also introduces the multiscale cascade module and the channel attention mechanism. This article conducted ablation study and qualitative and quantitative comparisons on public datasets, and the results show that the application of AGA significantly compresses the model size while ensuring performance, and other application components have the significant gain on the network. Compared with other advanced methods, the network in this article has outstanding performance.
engineering, electrical & electronic,instruments & instrumentation
What problem does this paper attempt to address?