An Adaptive Underwater Image Enhancement Framework via Multidomain Fusion and Color Compensation
Yumeng Tian, Kangchen Yao, Xiaoyang Yu, Tian Zhou, Sen Xu · IEEE Sensors Journal · 2025
Underwater image enhancement is essential for various applications, including marine exploration, ecological monitoring, and robotic vision. However, the complex absorption and scattering characteristics of the underwater environment often cause severe color distortion, contrast degradation, and visual noise, particularly under varying illumination, turbidity, and water conditions. To address these challenges, we propose an Adaptive Underwater Image Enhancement Framework (AUIEF) that is both physically inspired and perceptually guided. The framework consists of three dedicated stages: a preprocessing module that employs hybrid multi-scale illumination normalization and wavelet-guided denoising to handle non-uniform lighting and spatial noise; an Adaptive Underwater Dehaze Compensation (AUDC) module that recovers wavelength-dependent color attenuation through spectral absorption modeling; and a Perceptual-Driven Color Balance (PDCB) module that optimizes color presentation using the CIEDE2000 metric to align with human visual perception. Notably, the proposed framework operates in a training-free manner without requiring any pre-trained datasets, ensuring computational efficiency and strong adaptability to diverse real-time underwater scenarios. Extensive experiments on two benchmark datasets (UIEBD and U45) demonstrate that AUIEF achieves state-of-the-art performance in visibility enhancement, noise suppression, and color fidelity, validating its effectiveness and practical value for underwater imaging applications.