Geometric based structure propagation and texture matching for 3D texture completion

This paper proposes a novel method to complete the missing appearance of mesh surfaces through structure propagation and texture matching. In advance, the appearance on surfaces is divided into structure and texture components. First, the missing structures are predicted through structure propagation. The structure propagation is guided by surfaces geometric features by using 3D tensor voting and crest line detection. Then, with the guidance of predicted structures, the optimal local sample patches are matched to inpaint the missing texture. Finally, the completed result is obtained by mapping the optimal patches to the missing texture regions on surface. We acquire convincing completion results as shown in the experiments and comparisons.