An Improved SwinIR Preprocessing Network for High-Precision Digital Image Correlation Measurements

To address noise interference, limited resolution, and speckle image quality degradation in Digital Image Correlation (DIC) measurements under complex imaging conditions, this paper proposes a speckle image preprocessing method, termed Speckle-SwinIR, for high-precision DIC measurements. Based on the SwinIR architecture, a Speckle Decorrelation Compensation (SDC) module is introduced to restore decor-related textures using speckle-granularity priors and enhance speckle consistency. A Cross-scale Deformable Attention (CDA) mechanism is employed to replace the fixed-window attention scheme for modeling cross-scale feature variations in complex speckle textures. The proposed framework performs denoising and ×2 super-resolution reconstruction on low-quality speckle images, thereby enhancing the representation of speckle detail for subsequent sub-pixel DIC displacement measurements. A hybrid dataset consisting of simulated and real speckle images is constructed, and degraded sample pairs are generated via noise injection and resolution reduction to evaluate image restoration quality and its contribution to improved DIC accuracy. The proposed method is further applied to a high-speed rock three-point bending splitting experiment. Experimental results show that the proposed method improves image quality, reduces DIC displacement measurement errors, and enhances the continuity and smoothness of the displacement field.