Accelerating digital volume correlation computation using optimized sub-volume downsampling

Digital volume correlation (DVC) is a powerful yet practical image-based experimental technique for extracting internal full-field three-dimensional (3D) deformation. However, its practical application is severely hindered by high computational cost, primarily due to the large number of voxel points within each cubic sub-volume. During the implementation of the state-of-the-art DVC algorithm, sub-voxel intensity interpolation must be performed for every voxel point in each iteration, which results in a huge computational burden and greatly decreases the overall computation efficiency. However, not all voxel points within a sub-volume contribute meaningfully to the matching accuracy and precision. Based on this finding, a novel optimized sub-volume downsampling strategy is proposed. By selectively retaining only key voxel points with significant influence on matching quality, this strategy accelerates DVC computation without sacrificing sub-voxel registration precision. The strategy introduces a measure called mean intensity gradient (MIG_mean) for quantifying the significance of each voxel point within sub-volumes. An adaptive threshold determination approach is developed to tailor the MIG_mean threshold for each sub-volume based on its internal intensity variation. Only voxel points with MIG_mean values exceeding the adaptive threshold are retained for correlation analysis, thereby reducing the number of points involved in intensity interpolation while preserving essential intensity gradient information for accurate and precise DVC measurement. The effectiveness and practicality of the method are validated through both simulation and real tests. Experimental results show that the proposed strategy achieves up to an 80 % reduction in per-point computation time and a nearly 5 times improvement in computational efficiency, while maintaining comparable measurement accuracy and precision to routine DVC calculation.