Ghost imaging in dynamic underwater environments by dual-measurement intensity correction

A major obstacle to underwater imaging is water-column scattering, especially under dynamic or turbulent conditions. Within the framework of computational ghost imaging, the inherent averaging effect partly counteracts this scattering, and according to the central limit theorem as well as Monte-Carlo studies, the resulting detection noise tends toward a Gaussian distribution. Although some deep-learning approaches train on synthetically added white Gaussian noise, real underwater scenes almost never provide clean ground-truth images, so the final result critically depends on the initial detected intensities. To address this, we present an intensity-correction protocol: two independent measurements are averaged, their noise variance is estimated on-the-fly, and a diagonal Wiener filter is applied to correct the intensities. Simulations and experiments both show that the method markedly improves the reconstruction quality of underwater computational ghost imaging.