Abstract:
Model acquisition process usually produce incomplete surfaces due to the technical constrains. This thesis presents the algorithm to perform surface completion using the available surfaces context. Previous works on surface completions do not handle surfaces with near-regular pattern or irregular patterns well. The main goal of this research is to synthesize surface for hole that will have similar surfaces context or geometric details as the holes surrounding. The surfaces can have near-regular patterns, irregular patterns or stochastic patterns. This research uses multi-resolution approach to decompose the model into low-frequency part and high-frequency part. The low-frequency part is filled smoothly. The high-frequency part are transformed it into the Laplacian coordinate and filled using example-based synthesize approach. Laplacian coordinate, which encodes normal and curvature of each point on the surface, is used as the surface signature when perform geometric detail similarity search. After all the two resolution parts is filled, they are combined in Laplacian coordinate and are transformed using inverse Laplacian transform to reconstruct the complete surface. The algorithm is tested with planar surfaces and curve surfaces with all kind of relief patterns. The results indicate that the holes can be completed with the geometric detail similar to the surrounding surface. The seam between the hole and the surrounding surface is hardly visible.