Underwater visual image enhancement method based on the fusion of optical-model methods and model-free methods

The ability of underwater robots to autonomously identify specific targets is fundamental for underwater robots to complete underwater tasks autonomously. However, the quality of underwater images decays with increasing underwater depth, resulting in low-contrast and high color distortion. To address this issue, an underwater image enhancement algorithm based on the fusion algorithms of optical-model methods and model-free methods is proposed in this work. The depth map features of a specific scene are first acquired, and the depth map is used as input to measure the attenuation and transmission maps. Subsequently, the images were optimized at the pixel level by improving the model-free based methods of background light estimation, improving the red channel prior, improving gamma correction, and image sharpening. For underwater images with light sources and white bodies, we use a new estimation algorithm based on global background illumination in the background light estimation part, which successfully solves the problem of uneven illumination caused by traditional processing. For the red channel prior, we proposed an adaptive gamma correction method. We connect image pixels with gamma values using the region mapping technique to improve the image quality without knowing the illumination information. Finally, a novel pyramid fusion technique is proposed to fuse the results of optical-model methods and model-free methods. In this paper, the UIQM and UCIQE are used as underwater image quality evaluation metrics. Experimental results show that the proposed method improves 2.23 and 1.311 in two metrics compared to the FUnlE-GAN algorithm, 0.34 and 1.331 compared to the Deep WaveNet algorithm, 0.45 and 1.241 compared to the Retinex-based algorithm, 0.78 and 1 compared to the Automatic red channel algorithm, 0.21 and 0.699 compared to the WCID algorithm, respectively.