HG2former: HSV-Gamma Guided Transformers for Efficient Underwater Image Enhancement

Abstract

Due to optical phenomena, such as the absorption and scattering of light in underwater environments, underwater images often suffer from degradation in color, contrast, clarity, and noise. Existing deep learning-based methods for underwater image enhancement typically learn a direct mapping from low-quality to high-quality underwater images, without fully considering the mapping of local luminance, chrominance, and contrast features. In this article, we propose a transformer model guided hue, saturation, value (HSV) and gamma correction for underwater image enhancement. The HG2former combines the HSV color model and gamma correction techniques to isolate the three fundamental characteristics of color, providing rich, differentiated enhancement for both color and contrast in underwater images. In addition, nonlinear gamma correction adaptively adjusts the brightness and contrast of images, addressing issues of visibility reduction and color distortion in underwater imaging. Furthermore, we introduce a meticulously designed encoder–decoder structure, along with an improved multihead self-attention module, to capture the spatial distribution patterns of underwater images while modeling both local and long-range dependencies. Extensive experimental results on multiple data sets demonstrate that the proposed HG2former outperforms other state-of-the-art methods.