CRCNet: Underwater image enhancement via color adversarial empowered residual collaborative network

The challenging underwater environment and the attenuation of light during propagation often lead to significant image degradation, including light absorption, scattering, color distortion, and loss of fine details. To overcome these challenges, we propose an underwater image enhancement network based on a color-adversarial empowered residual collaborative network, termed CRCNet. CRCNet adopts a dual-branch architecture consisting of a main branch for global image enhancement and a sub-branch for color compensation. The main branch incorporates a multi-scale collaborative enhancement module into a U-Net framework guided by hybrid attention mechanisms, enabling effective enhancement of image clarity while preserving structural and textural details. Meanwhile, the sub-branch introduces a Color Adversarial Compensation Module, which integrates traditional color correction priors to alleviate common color shifts and distortions in underwater scenes. Through a collaborative feature-level fusion strategy, the two branches jointly improve color naturalness and structural fidelity, achieving a better balance between image sharpness and color accuracy. Extensive experiments on multiple public underwater image datasets demonstrate that CRCNet consistently outperforms state-of-the-art methods. In particular, it achieves a PSNR of 26.458 on the Large-Scale Underwater Image Dataset (LSUI), validating its strong generalization capability for real-world underwater imaging applications.