網格分割轉換

Abstract

我們提出了一個網格分割轉換演算法，將來源網格的分割結果一致地對應到目標網格上。首先我們提出一個新的骨架對應方法，可以同時對應來源網格和目標網格之間的骨架節點以及連結，接著在來源網格上將每個分割邊緣線對應到所屬的骨架連結，最後將每個分割邊緣線轉換到目標網格所對應的骨架連結上。我們的骨架對應演算法對骨架的連接性較不敏感，而且利用了骨架連結的語義性群聚以及空間配置的資訊。我們的分割轉換演算法致力於在目標網格上產生分割邊緣線，使其特徵以及局部幾何細節和來源網格對應的分割邊緣線能夠一致。最後我們在各種類型的物體上展示了多樣的分割轉換結果，並和目前最新的相關研究作比較。

We present a novel segmentation transfer that maps consistent segmentation of a source mesh to a target mesh. We first propose a skeleton correspondence method to match the skeleton links and nodes between two skeletons of the source and target meshes, then associate each segment boundary with a skeleton link of the source mesh, and finally map each boundary to the corresponding skeleton link of the target mesh. Our skeleton correspondence computation is less sensitive to the skeleton connectivity and employs the clustering of semantically similar links and spatial information. The boundary transfer procedure aims to produce the segment boundaries of the target mesh whose locations and geometric properties are consistent to the ones of source. We demonstrate the segmentation transfer results on various meshes and compare with the state-of-the-art techniques.