Underwater Image Enhancement using Scene-Specific Red Channel Prior and Fusion

Abstract

Quality of underwater images are reduced with the depth since light is absorbed and scattered by the suspended particles in water. Although several optical model-based and model-free underwater image enhancement methods have been developed in the past, their performances are considerably poor since they fail to consider the scene-specific cues and depth information. In this study, a novel enhancement approach is proposed by combining optical model-based and model-free algorithms. In the model-based technique, a tentative depth map estimation technique is proposed and then used to compute more accurate red channel prior, background light, and transmission. White balancing and Gamma correction are used in the model-free technique. Outputs of both techniques have been combined by a fusion strategy to recover the details and colour. Experimental results demonstrate the proposed enhancement method outperforms state-of-the-art approaches on UIEB, OceanDark, and U45 benchmark datasets.