Background <p>Digital image correlation (DIC) enables full-field deformation measurement. Accurate measurement in irregular boundary regions is particularly important, since stress concentration and fracture often originate in these areas.</p> Objective <p>This study systematically evaluates the boundary measurement performance of three representative DIC frameworks—local Subset-DIC, global Mesh-DIC, and physics-informed neural network-based DIC (PINN-DIC)—to clarify their applicability in complex boundary conditions.</p> Methods <p>Simulated deformation fields with quadratic and complex curved boundaries, together with experimental data from radial compression fracture tests on ring specimens, were used to assess displacement and strain accuracy as well as noise robustness in boundary regions.</p> Results <p>PINN-DIC maintains high accuracy, smoothness, and robustness under complex boundary conditions. Mesh-DIC performs well for quadratic curved boundaries but is sensitive to mesh discretization in irregular geometries. Subset-DIC exhibits displacement discontinuities and relatively large errors near boundaries.</p> Conclusions <p>Benefiting from global and strongly nonlinear fitting, PINN-DIC demonstrates superior adaptability and broader application potential for deformation measurement in irregular boundary scenarios. Nevertheless, further refinement is needed to better regulate the global parameter coupling effects in highly localized deformation regions.</p>