Underwater Image Enhancement Based on Light Attenuation Prior and Dual-Image Multi-Scale Fusion
Liushanchuan He, Yongjie Yan, Chengtao Cai · 2022 Global Conference on Robotics, Artificial Intelligence and Information Technology (GCRAIT) · 2022
The focus of this work is solving the degradation of underwater images. Different from images taken on land, underwater images suffer from color bias and fuzzy details due to absorption and scattering. In order to solve the above problems of underwater images, we propose a novel method called LAPF, which includes two stages of restoration and dual-image multi-scale fusion. Restoration based on underwater light attenuation prior can initially restore the details of the underwater image by building a scene depth estimation model. Coupled with the use of image fusion, obtain the contrast-enhanced picture and color-corrected picture as fusion inputs, and calculate the normalized weight maps to add detail. Carrying out multi-scale decomposition and fusion of these two parts, successfully further enhance the contrast while realizing the color correction. Experimental results on a variety of underwater images show the power of the method proposed in this paper, which achieves good results in both subjective visual perception and objective quality evaluation parameters.