| Pose transfer means that transfer the pose of target character to another character.By studying pose,artists can reuse the existed character.Compared with usual deformation method,pose study only needs few users’ interaction and simplifies art creation,which makes it in great need in manga industry and movie industry.The core idea of pose study is to make original character learning the pose of target character while preserving original character’s geometry feature.Because the pose study method on three dimensional graph can’t be applied in image and 2-dimensional character,our work proposes the 2-dimensional pose study framework based on multi-resolution decomposition.The framework use Laplace operator to divide character into high resolution part which contains detail information and low resolution level which contains body.We make body transfer pose and reconstruct with detail information to get the final result.In traditional method,preserving detail and transferring pose are in one step,which make user difficult to control the division of detail and pose.Our method let user determine detail and pose and increase the controllability of pose study process.Then,we erase the high frequency information when transferring pose on character body.Compared with traditional method,we have privilege in separating and preserving the somatotype of body.In addition,we divide the high resolution detail information in advance;compared with traditional linear adding method,we can implement the conformal transformation when reconstructing the character.When transferring pose between bodies,we have proposed two innovative methods.One is dealing graph based on space domain,we proposed skeleton skinning pose transfer algorithm.The advantages of our algorithm are in well preserving the somatotype information of character and being suitable to the situation that two characters’ poses differ much with each other.The other method is to deal with graph in spectral domain,we proposed pose transfer algorithm by signal substituting.It has solved the problem faced by current pose transfer method,which is the uncouple problem of graph’s basis when processing the graph with geometrical information involved Laplace operator.Our signal transfer algorithm gets good pose transfer result compared with current algorithm.In our proposed two pose transfer algorithms,geometrical constraints are adopted to preserve graph’s own geometrical properties.In the skeleton skinning pose transfer algorithm,the constraints refine the pose transfer result.In the pose transfer algorithm by signal substituting method,besides refining result,the constraints are also used to drive the movement of inner part with the low resolution outline.Sufficient experiments have been done on our two pose transfer algorithm to illustrate its effectiveness,robustness,and convergence.Experiment results show that our method can successfully transfer the pose of 2-dimensional characters. |
