Pre‐Trained Codebook‐Based Enhancement: A Novel Approach for Clarifying Underwater Images

ABSTRACT Underwater images often suffer from degradation due to light attenuation, scattering and noise, leading to colour distortion and loss of detail. Traditional enhancement methods struggle to address these issues effectively, particularly due to the scarcity of paired underwater datasets. This paper introduces a novel underwater image enhancement method leveraging a pre‐trained codebook to utilize scarce paired data effectively. Our approach employs a multi‐channel encoder with an attention mechanism to extract multi‐scale features, ensuring alignment with reference features from a pre‐trained VQGAN encoder. A multi‐feature fusion decoder integrates shallow and deep features to mitigate spatial information loss, while a gradient estimation branch enhances edge and high‐frequency details. Experimental results demonstrate that our method outperforms existing techniques in terms of contrast enhancement (PSNR: 23.52), detail restoration and clarity improvement, significantly enhancing the visual perception of underwater images. The source code will be made available upon acceptance.