Multiresolution Mean Shift Clustering Algorithm for Shape Interpolation
Hung-Kuo Chu,Tong-Yee Lee
DOI: https://doi.org/10.1109/tvcg.2009.40
IF: 5.2
2009-01-01
IEEE Transactions on Visualization and Computer Graphics
Abstract:In this paper, we solve the problem of 3D shape interpolation with significant pose variation. For an ideal 3D shape interpolation, especially the articulated model, the shape should follow the movement of the underlying articulated structure and be transformed in a way that is as rigid as possible. Given input shapes with compatible connectivity, we propose a novel multiresolution mean shift (MMS) clustering algorithm to automatically extract their near-rigid components. Then, by building the hierarchical relationship among extracted components, we compute a common articulated structure for these input shapes. With the aid of this articulated structure, we solve the shape interpolation by combining 1) a global pose interpolation of near-rigid components from the source shape to the target shape with 2) a local gradient field interpolation for each pair of components, followed by solving a Poisson equation in order to reconstruct an interpolated shape. As a result, an aesthetically pleasing shape interpolation can be generated, with even the poses of shapes varying significantly. In contrast to a recent state-of-the-art work (Kilian et al., 2007), the proposed approach can achieve comparable or even better results and have better computational efficiency as well.