Reinforcement Learning-Driven Framework for Robust Underwater Image Enhancement

Underwater imaging poses persistent challenges such as color distortion, low contrast, and image noise, primarily due to light absorption and scattering in aquatic environments. This paper proposes a robust image enhancement pipeline composed of traditional image processing techniques—white balance, CLAHE, wavelet denoising, guided filtering, and Multi-Scale Retinex. To improve adaptability under diverse underwater conditions, reinforcement learning (RL) is employed to automatically optimize critical enhancement parameters, guided by no-reference image quality metrics (UIQM and UCIQE). This solution balances perceptual performance and computational efficiency while remaining interpretable and lightweight. Experimental evaluation on the UIEB dataset demonstrates that the RL-optimized pipeline outperforms fixed-parameter baselines and achieves consistent improvements in perceptual quality. The framework is well-suited for real-time and resource-constrained underwater imaging applications.