Publication

Abstract : Accurate stress-field reconstruction in digital photoelasticity is based on high-resolution polarization images; however, polarization cameras inherently suffer from resolution limitations that degrade fringe continuity and reduce the accuracy of isochromatic and isoclinic extraction. Conventional interpolation techniques provide only marginal improvement, and existing deep learning–based super-resolution methods primarily optimize pixel fidelity without explicitly preserving the underlying physical structure of stress fields. We propose a structure-preserving super-resolution framework specifically designed for photoelastic polarization imaging. By incorporating a physics-consistent regularization inspired by elasticity theory, the proposed model enforces Laplacian and biharmonic consistency between reconstructed and reference fields, thereby maintaining curvature continuity and suppressing non-physical artifacts. Unlike prior divide-and-conquer strategies that require multiple degradation-specific networks, our approach employs a single lightweight model with fewer than one million parameters. The method demonstrates strong generalization to previously unseen geometries and maintains tight agreement with ground-truth stress distributions. This work establishes a physically informed super-resolution for polarization imaging and advances the practical reliability of digital photoelasticity in experimental mechanics and optical stress analysis.