- Conference date: 18–20 December 2012
- Location: Palm Garden Hotel, Putrajaya, Malaysia
Recently, applications of open surfaces in 3D have emerged to be an interesting research topic due to the popularity of range cameras such as the Microsoft Kinect. However, surface meshes representing such open surfaces are often corrupted with noises especially at the boundary. Such deformity needs to be treated to facilitate further applications such as texture mapping and zippering of multiple open surface meshes. Conventional methods perform denoising by removing components with high frequencies, thus smoothing the boundaries. However, this may result in loss of information, as not all high frequency transitions at the boundaries correspond to noises. To overcome such shortcoming, we propose a combination of local information and geometric features to single out the noises or unusual vertices at the mesh boundaries. The local shape of the selected mesh boundaries regions, characterized by the mean curvature value, is compared with that of the neighbouring interior region. The neighbouring interior region is chosen such that it is the closest to the corresponding boundary region, while curvature evaluation is independent of the boundary. The smoothing processing is done via Laplacian smoothing with our modified weights to reduce boundary shrinkage. The evaluation of the algorithm is done by noisy meshes generated from controlled model clean meshes. The Hausdorff distance is used as the measurement between the meshes. We show that our method produces better results than conventional smoothing of the whole boundary loop.
MOST READ THIS MONTH
MOST CITED THIS MONTH
Article metrics loading...