Abstract
Underwater image enhancement is a vital task in computer vision due to severe image degradation caused by wavelength-dependent light absorption and scattering in aquatic environments. This paper presents a novel enhancement framework, AURIE (Adaptive Underwater Retinex Image Enhancement), which integrates Retinex theory with adaptive color correction and frequency-based decomposition for robust underwater image restoration. The proposed method decomposes the input into low- and high-frequency components, enabling illumination correction in the HSI color space and detail refinement via edge-preserving filters. A customized RetinexNet is introduced to perform gamma-based illumination adjustment and feature-preserving reflectance enhancement using attention-guided refinement. The network is optimized using perceptual, structural, and color consistency losses. Extensive experiments on benchmark datasets (UIEB, EUVP) demonstrate superior performance in UCIQE, UIQM, PSNR, and SSIM compared to existing techniques. The architecture is also lightweight and computationally optimized, making it suitable for real-time deployment in autonomous underwater systems for applications in marine exploration and underwater robotics.
KEYWORDS
Underwater Image Enhancement (UIE), Retinex Theory, HSI Color Space, Image Decomposition, Illumination Correction, Retinexnet.
Palem Narasimhulu1*, Vishnu Soni2, Shivam Yadav3
1Research Scholar, APEX University, Jaipur-303002, Rajasthan, India
2Assistant Professor, APEX University, Jaipur-303002, Rajasthan, India3Research Scholar, APEX University, Jaipur-303002, Rajasthan, India