Hybrid Enhanced Underwater Dark Channel Prior Framework with Vision Transformer Refinement, Mamba-Based Fusion, and Diffusion-Driven Dehazing

Abstract

Underwater imaging can often exhibit colour casts, reduced contrast, and scattering effects due to wavelength-dependent absorption and light turbidity. This paper presents an enhanced underwater image restoration technique called Enhanced Underwater Dark Channel Prior with Advanced Refinement (EUDCP-AR), which combines a physics-based dehazing approach with modern refinement mechanisms for effective visibility recovery. The proposed framework incorporates a variation of Underwater Dark Channel Prior (UDCP) to determine the intensity of initial haze, a Vision Refinement with global attention based on transformer (ViT) to adjust local differences in the dark channel. A Mamba-fusion approach which is inspired by state-space refinements helps to improve the atmospheric light estimation through bidirectional brightness propagation, which results in better color balance and more uniform lighting.Subsequently, Diffusion helps in dehazing the transmission by way of a diffusion process, mapping and maintain edge information. Quantitative and qualitative experiments were done on various underwater datasets, measures of PSNR, SSIM, MSE and entropy as measures to evaluate the performance. The experimental findings prove that EUDCP-AR has better contrast enhancement, color fidelity and structural clarity with conventional and deep learning-based efficient methods of underground improvement. The reconstructed images stabilise natural tone, better contour, and definition of edges and are natural, noise artefacts were reduced, which proved the soundness of the hybrid physical-learning model. The proposed EUDCP-AR framework showcases a more robust, perceptually consistent and computationally efficient solution which helps in enhancing the underwater images. Its ability to preserve the fine details, balancing color attenuation and restoring the structural integrity, makes it good for applications in marine research, autonomous underwater vehicles (AUVs), submersible robotics, and scientific imaging.