透過您的圖書館登入
IP:18.191.181.231
  • 期刊

以局部傅立葉特徵為基礎之三維模型檢索系統

3D Model Retrieval based on Local Discrete Fourier Transform

摘要


隨著資訊科技的高速發展,電腦的軟硬體在效能上都有飛躍般的提高,使得三維模型可以被更精確的掃描與應用,例如電影阿凡達中的動態捕捉技術甚至可以即時捕捉三維座標的動態資訊來建立三維人物模型。隨著三維印表機的興起,使用者相較於以往可以利用簡單的器具將實體物件經過掃描來取得該物件的三維模型。面臨這些多樣化的模型與日益進步的三維技術,我們期望能夠使用一個較精確的檢索方法來輔助使用者取得需要的三維模型。在本論文中提出以局部特徵為基礎之三維模型檢索系統。為了更有效的擷取有代表性的特徵點,首先利用刪除三角片的方式將三維模型進行維度降低,接著以模型中心點來和所有座標與前者的距離來擷取出具有代表性的特徵點,然後對模型進行網格主軸分析演算法(Grid-based PCA, GPCA)來擺正模型。根據取得的特徵點以其為中心來對擺正模型擷取局部區域,稱為局部區塊。針對這些局部區塊,我們將使用三維傅立葉特徵(3D Discrete Fourier Transform)來取得區塊中的局部特徵。進行搜尋時,將利用局部特徵向量在資料庫中尋找使用者想要搜尋的三維模型,把相似度較高的三維模型回應給使用者。實驗中會採用五種資料庫來驗證三維模型檢索系統的效能,分別為3D Shape Retrieval Contest 2011(SHREC 2011)、3D Shape Retrieval Contest 2010(SHREC 2010)、3D Shape Retrieval Contest Watertight(SHREC-W)、普林斯頓(Princeton Shape Benchmark)、Purdue Engineering Shape Benchmark(ESB)。

並列摘要


With the pervasiveness of the 3D printer, a large amount of 3D models can be accessed through the Internet. The primary challenge to a 3D model retrieval system is to extract a set of proper features for efficiently representing and effectively discriminating distinct types of 3D models. In this thesis, a 3D model retrieval system will be proposed based on the local features. First, the resolution the 3D model will be reduced. The salient points of 3D model must be detected. Then, the grid-based principal component analysis (GPCA) is used to align 3D models. For each salient points, the 3D Discrete Fourier Transform (3D-DFT) is extracted to describe the 3D models. Experiments will be conducted on the Shape Retrieval Contest Watertight (SHREC'W), SHREC'10, SHREC'11, Princeton Shape Benchmark (PSB), and Purdue Engineering Shape Benchmark (ESB).

延伸閱讀