UWLDM: Underwater Image Enhancement Based on Latent Diffusion Model

Underwater images are often frequently impacted by complex environmental factors like water quality and uneven lighting, resulting in color bias and blurring. Existing methods via denosing diffusion probabilistic model (DDPM) generally lack flexible control mechanisms and suffer from long inference time, leading to suboptimal image enhancement performance and efficiency. To address these issues, we propose a novel underwater image enhancement (UIE) method combining the latent diffusion model (LDM) and attention mechanism, abbreviated as UWLDM. First, UWLDM adopts an encoder to project the reference images from pixel space to a lower dimensional latent space. Secondly, we perform the diffusion process to extract the high-level structural information from the images with various scenes and learn their latent representations. Finally, we employ a decoder to realize image restoration where the degraded images are its conditional guidance. We evaluated UWLDM on the three UIE datasets and compared it with eight state-of-the-art models. Experimental results demonstrate that UWLDM effectiveness surpasses current UIE methods, exhibiting better generalization ability and roubustness.