- مبلغ: ۸۶,۰۰۰ تومان
- مبلغ: ۹۱,۰۰۰ تومان
In spectral decomposition of a 3D mesh model, it is well known that eigenvectors with respect to small eigenvalues determine its main pose while eigenvectors associated with large eigenvalues encode its surface details. Based on this property, given two meshes with different connectivities, one can use coupled quasi-harmonic technique to transfer the pose of one mesh onto the other by exchanging the low-frequency coefficients in their spectral representations. However, directly synthesizing the new low frequencies with old high frequencies usually exhibits two vital artifacts: one is detail shearing and shape collapsing, and the other is medium-scale pose missing. This paper reformulates the pose transfer as a deformation problem with low-frequency coefficients as handles. It finally leads to a non-linear optimization with the coefficients as data constraint and Laplacian coordinates as regularity term for preserving details. Meanwhile, a hierarchical pose transfer framework is introduced to capture the medium-frequency poses. To reduce the computation complexity and enhance the stability we further solve the problem in a subspace defined by mean-value coordinates.
We have described a spectral-based pose transfer approach. It is insufficient to learn the pose of a reference by only using low-frequency component. Therefore, our approach firstly creates a hierarchical pose structure to capture mediumscale poses. It then segments the components including medium-scale pose. Finally, the approach performs pose transfer operation for low-scale frequency on two segmented parts. To prevent distortion during transferring process, we introduce a penalty to preserve Laplacian coordinates. Based on the new deformation formulation, a framework is established for a variety of applications such as pose transfer between two meshes with different connectivities, mesh deformation, and shape interpolation.As future work, the current computational method for coupled quasi-harmonic basis is not stable enough, particularly, when one of the mesh models is too coarse or too irregular. In addition, at this stage, we need to interactively specify parts in which pose learning is incomplete. It is desirable to automatically detecting and segmenting those parts. Besides, triangulation quality of cages is also a significant factor impacting on the transferring results. Adaptively refining cages may be a good choice to control the degree of freedom. Finally, how to automatically choose the weights in Eq. (12) is an important issue we need to tackle in the next step.