<p>Underwater images suffer from low contrast, blur, and color deviation due to light absorption, scattering, and distortion. Traditional restoration methods, encompassing physical model-based compensations and basic filtering techniques, often struggle to adequately address these complex degradations. While deep learning approaches have demonstrated promising results on specific datasets, their ability to adapt to new conditions and operate reliably across varying illumination conditions and complex underwater scenarios remains challenging. To tackle these limitations effectively, this paper proposes a novel deep learning framework, termed WCM-Net, which incorporates wavelet transform, color compensation module, and dedicated multi-scale feature extraction block to decouple and enhance latent information at multiple frequency bands and spatial resolutions. ​​This integrated architecture efficiently learns comprehensive global-to-local feature representations for underwater image restoration. Experimental results on multiple public underwater datasets demonstrate that the proposed WCM-Net effectively enhances visual quality and significantly improves performance in downstream tasks such as key point detection and edge processing. Compared to existing state-of-the-art techniques, our approach exhibits superior robustness and adaptability, offering a valuable solution for underwater image processing applications.</p>