Post-registration three-dimensional intensity compensation for precise correction of wood’s anisotropic shrinkage in time-resolved computer tomography (CT) imaging

X-ray computed tomography (CT) tracks structural changes in wood during drying, but anisotropic shrinkage and complex anatomy cause voxel misalignments, compromising density-based analysis. Two-dimensional (2D) registration methods assume uniform material behaviour and perform poorly in heterogeneous fibre regions. This study introduced a three-dimensional (3D) density-compensation method that combines non-rigid 3D registration, continuous deformation-field estimation, and volumetric mass-conserving redistribution, using displacement fields from anisotropic dimensional changes to ensure accurate, time-consistent voxel-wise density estimation. Time-resolved CT scans of specimens with and without knots (regions with fibre disturbances) were used to assess the method’s performance. The 3D method substantially improved voxel alignment and topological consistency, particularly in knotty areas where 2D approaches perform poorly. The deformation field remained physically valid (mean Jacobian > 0.95), with mean volumetric shrinkage around 10%, consistent with Norway spruce in this moisture range. Registration metrics improved with the structural similarity index measure (SSIM) from 0.62 to 0.92 and the normalised mutual information (NMI) from 0.52 to 0.85. In complex regions such as knots, 2D and 3D estimates differed by more than ±40 kg/m3, with local deviations up to ±200 kg/m3. While 2D compensation may suffice in homogeneous wood, 3D compensation is necessary for reliable density analysis in anatomically complex regions.