AI-driven fringe restoration for overexposed condition using a GAN-based framework

Fringe projection profilometry (FPP) is prone to overexposure when scanning highly reflective surfaces, causing fringe saturation and loss of phase information. We propose FDCGAN, a frequency-domain constrained GAN that restores saturated fringe patterns by designing frequency-aware networks with Fourier-based loss functions. Trained on synthetic data and fine-tuned with real measurements, FDCGAN achieves superior performance in fringe recovery and 3D reconstruction under extreme exposure. The method is practical for industrial and biomedical FPP applications facing challenging lighting conditions.