多邊形網格分割轉換

Abstract

物體網格分割(mesh segmentation )為一重要的三維模型技術，為許多圖學演算法與應用的前置處理，如碰撞偵測、網格參數化、物體骨架、紙模型切割等等。 目前的全自動物體網格分割演算法有諸多的問題，例如，類似的物體分割結果可能不一致，分割的結果也常不是使用者所預期。我們將發展一個創新的全自動物體網格分割演算法，稱為物體網格分割轉換(segmentation transfer)，對使用者提供的一個已經分割完成的範例物體，將此分割的樣式轉換到另一個目標物體上，使兩個物體擁有一致的物體分割結果。本計畫預期將提出一個以物體外型對應（shape correspondence）為基礎的網格分割轉換系統。而物體外型的對應將以骨架對應為基礎，透過參數化及其他骨架對應之表面幾何資料而得。本計畫將首先發展 骨架對應的方法，期待此結果能比過去的方法快而且可靠。然後，以骨架對應的結果發展物體之外形對應，並以此對應將範例物體上的分割轉換到目標物體上。

Mesh segmentation is a fundamental modeling technique that usually serves as a preprocessing for collision detection, skeletonization, parameterization, and paper crafting. Current mesh segmentation algorithms suffer from several problems, such as the inconsistent segmentations for objects that are similar or in different poses and segmentation results that are not what users expect. In this project, we will develop a new mesh segmentation scheme, called segmentation transfer, aiming to transfer the segmentation on a source object to a target object such that these two objects have the similar segmentation. The segmentation transfer algorithm will be based on shape correspondence between the source and target objects. We will first study the skeleton correspondence problem and then develop the shape correspondence based on a mesh parameterization that also possesses a mapping to the skeleton. Based on the shape correspondence, a fuzzy region on the target object that is corresponding to a segmentation boundary curve on the source object will be derived and within which a segmentation boundary curve can be derived by using the graph cut algorithm.