Bo Tang

High-fidelity 3D profilometry of non-Lambertian surfaces is severely degraded by the co-occurrence of specular saturation (clipping harmonics) and global illumination (GI). Although Micro Phase Shifting (MPS) isolates low-frequency GI via narrowband projection, non-sinusoidal saturation clipping invalidates its fundamental inter-frequency amplitude consistency constraint, causing phase unwrapping collapse. To resolve this coupled dilemma without exceeding the theoretical minimum image count ($F+2$ frames), this paper proposes a physics-guided adaptive spectral decoupling framework. We demonstrate that the inherent narrowband spatial carrier of MPS establishes an analytical passband, enabling selective harmonic elimination while strictly preserving GI resilience. An adaptive Butterworth low-pass filter integrated with boundary-symmetric mirror extension is formulated, where the optimal cutoff radius is dynamically locked via a spatial correlation convergence criterion. This restores fundamental sinusoidal fidelity, reinstates MPS matrix amplitude consistency, and eliminates Gibbs boundary ringing. Simulations and physical experiments on specular metallic specimens show that our method achieves up to an 87.5% reduction in Root Mean Square Error (RMSE) and a 60.9% reduction in Mean Absolute Error (MAE) over standard MPS, while outperforming multi-exposure fusion benchmarks using 74.1% fewer images. Crucially, high metrological precision is achieved without compromising sharp geometric step edges.