以骨架驅動之交互參數化

Abstract

我們提出一個以骨架資訊來產生兩三維物體間交互參數化之方法。給予兩個物體，我們擷取出其骨架並取得骨架與物體表面間、以及骨架與骨架間之對應資訊。兩物體接著會透過切割程序分成數個相對應的區塊，接著我們將每一個對應的區塊分別參數化到一個平面上。而每一參數化之分割區塊，我們則透過一個迭代放鬆法來產生分割區塊間之交互參數化，此迭代放鬆法主要由骨架與表面以及骨架與骨架間之對應資訊來驅動。兩物體分割區塊之交互參數化產生後，我們於分割區塊之邊緣處使用一個平滑化的程序來產生出最後的對應結果。概要來說，我們使用了密集的骨架與物體表面間之對應資訊當作軟性限制來建立交互參數化。相較於先前之技術，此方法可以自動的產生出物體間有意義之特徵對應且無須給予對應限制點，並僅需提供少量之對應點在分割區塊的邊緣處即可。此方法也可以自動的產生出特徵的部分對應。

We propose a novel skeleton-driven framework for cross-parameterization between two 3D polygonal models. Given two models, we first extract their curve skeletons, retrieve the skeleton-to-surface mapping information, and derive a skeleton-to-skeleton mapping. We then segment both meshes into several consistent parts and for each part pair we embed the parts in correspondence onto a planer domain. After the process of embedding, we apply an iterative relaxation scheme on each parameterized part that takes the skeleton-to-surface and skeleton-to-skeleton mapping information into account. A smoothing process is invoked at the end so as to smooth the cross-parameterization along the part boundaries. Briefly speaking, we use the dense skeleton-to-surface mapping information as the soft constraints during the cross-parameterization construction. As a result, our method can match the object features automatically without the general feature constraint points as common approaches. The method requires only a few user inputs that specify matching points on the part boundaries. We also show that our method can generate partial mapping results for the models that have unique object features.