Edge-enhanced real-time holography using physics-guided residual learning

Conventional double-phase holography for broadband images suffers from aliasing artifacts in high-frequency regions due to spectral overlap between signals and misalignment noise. We propose ResDPH-a deep learning-assisted double-phase encoding framework with residual compensation-that embeds a neural network within a physical model to enhance detail fidelity. Simulations show ResDPH generates 2 K holograms at 80 fps, achieving 3 dB higher average PSNR than baseline methods (peak: 32.51 dB). Optical experiments validate significant noise suppression and edge recovery.