Physics-driven saliency-weighted image enhancement for computational ghost imaging through turbid water

Imaging through strongly scattering media is hindered by diffuse background noise. We report saliency-weighted computational ghost imaging (SW-CGI), a non-iterative post-reconstruction enhancement strategy exploiting the statistical energy disparity between signal-dominated and scattering-background-dominated regions in CGI-reconstructed images. By applying an adaptive sigmoid-based soft-masking mechanism derived from intensity statistics, SW-CGI isolates the signal support while preserving essential "snake" photons. Experiments in turbid water (up to 80 NTU @ 15 cm) demonstrate a ∼6.7 dB PSNR gain over computational ghost imaging at 10 NTU, significantly outperforming standard spatial denoising in retaining high-frequency textural details.