Restoration of underwater optical image with non-uniform illumination via foreground and background segmentation

Optical images captured underwater often suffer from insufficient lighting, and artificial auxiliary light may produce non-uniform illumination, which will degrade image quality and hide valuable visual information. To overcome this limitation, we propose a novel foreground-background separation based methoddepending on an improved optical image formation model for underwater optical image restoration. Considering the color disparities between the foreground and background regions in underwater optical images, we first extract the background region by employing three simple but effective features including gradient, color difference, and area proportion. Also, we estimate the local background light and transmittance using underwater maximum reflectance prior. After that, we acquire the latent clean image by performing a weighted map between the restored foreground and enhanced background. Both the qualitative and quantitative comparisons demonstrate the superior performance of the proposed method. The code is available at: https://github.com/Hou-Guojia/FBS-UOI.