Underwater image super-resolution via multi-domain learning

Underwater images suffer from haze effects and low contrast due to wavelength- and distance-dependent scattering and attenuation. These issues present significant challenges for various underwater vision applications. Super resolution (SR) of underwater images offers an effective solution for enhancing both detail refinement and overall image visibility. However, underwater image SR remains challenging owing to the severe degradation of texture and color information. This paper proposes a multidomain learning-based SR network to enhance the performance of underwater image SR. Specifically, we introduce a multidomain encoder network that integrates grayscale and dual-color spaces into a unified framework. This architecture enables our model to simultaneously improve the underwater image quality through texture enhancement and color correction. By incorporating a channel attention mechanism, the most discriminative features extracted from multiple domains can be adaptively weighted and fused. Consequently, our network effectively boosts image resolution and enhances visual quality by leveraging multidomain data and the advantages of learning-based approaches. Experimental results demonstrate the superior performance of the proposed model in underwater image SR.