Underwater image restoration using Joint Local–Global Polarization Complementary Network

Abstract

Underwater image always suffers from the degradation of visual quality and lack of clear details caused by light scattering effect. Since polarization imaging can effectively eliminate the backscattering light, polarization-based methods become more attractive to restore the image, which utilize the difference of polarization characteristics to boost the restoration performance. In this paper, we propose an underwater image restoration using joint Local–Global Polarization Complementary Network, named LGPCNet, to achieve a clear underwater image from multi-polarization images. In particular, we design a local polarization complement module (LCM) to adaptively fuse complementary information of local regions from images with different polarization states. By incorporating this, we can restore rich details including color and texture from other polarimetric images. Then, to balance visual effects between images with different polarization states, we propose a global appearance sharing module (GSM) to obtain the consistent brightness across different polarization images. Finally, we adaptively aggregate the restored information from each polarization states to obtain a final clear image. Experiments on an extended natural underwater polarization image dataset demonstrate that our proposed method achieves superior image restoration performance in terms of color, brightness and contrast compared with state-of-the-art image restored methods.