Deep-Learning-Driven Dual-Channel Dynamic Digital Virtual Phase-Shifting Fresnel Incoherent Correlation Holography (D⁴FINCH)

Fresnel incoherent correlation holography (FINCH) enables scan-free 3D imaging with an incoherent light source. However, traditional reconstruction techniques rely on multi-step phase-shifting processes to suppress background noise and eliminate twin-image interference, which inherently limits temporal resolution. Here we present a deep-learning-driven, dual-channel dynamic FINCH framework with digital virtual phase shifting (D⁴FINCH). Using a dual-channel module, two fixed phase-shifted holograms are captured simultaneously in a single exposure, and a tailored network (H-Net) synthesizes the missing phase steps computationally, forming a “virtual four-step” set equivalent to conventional acquisition. Reconstruction experiments show that the virtually phase-shifted holograms closely match their multi-exposure counterparts. Quantitative evaluations (PSNR and SSIM) indicate reconstruction quality comparable to the four-step method, with effective suppression of background noise and twin-image interference. Multi-depth focusing experiments further verify faithful 3D reconstruction across axial planes. Since D⁴FINCH requires only a single exposure, it markedly improves temporal resolution while preserving high-fidelity reconstructions, making it well suited for real-time holography of dynamic scenes and providing a powerful and efficient approach toward real-time, high-quality 3D imaging in FINCH systems.