自组织映射网路及核算理论于隐函曲面重建implicit-先进工程学刊.pdfVIP

先進工程學刊第六卷 第二期 87 Journal of Advanced Engineering Vol. 6, No. 2, pp. 87-95 / April 2011 自組織映射網路及核算理論於隱函曲面重建 Implicit Surface Reconstruction Based on SOM Network and Kernel Method 1 *1 2 王中行 賴泰華 張庭瑞 Chung-Shing Wang1, Tai-Hua Lai*1, Teng-Ruey Chang2 摘要 諸多有關隱函曲面 (Implicit Surface)的文獻指出，透過徑向基函數 (Radial Basis Function, RBF) 核算 (Kernel Method) 來建立隱函曲面雖有許多好處，卻也存在某些限 制：龐大的樣本數除了需要大量的系統資源來存取矩陣資料外，大量的 RBF中心將造 成系統計算上嚴重之負擔，使得以該方法為基礎的隱函曲面建構論毫無用處。故本研究 運用自組織映射 (Self-Organizing Map, SOM) 網路結合核算理論，由快速重建為方向， 從事深入的理論探討及模擬，研擬的課題包括「 SOM 特徵擷取」及「RBF建面」等子 題。由研究結果可知，透過 SOM網路，我們可得到足以描述原模型幾何的特徵資料， 而核化的算則，則使得隱函曲面的計算更加簡單且有效率，也確實達到文中所預期的破 面修補效果。 關鍵詞 ：逆向工程，曲面重建，隱函曲面，類神經網路，核算法 Abstract The benefits of modeling implicit surface with RBF Kernel have been recognized by numerous bibliographies. Nonetheless, this work was restricted to small problems by the storage and arithmetic operations of direct method. When processing the RBF kernel estimate, to considerate whole Euclidean distance between RBF centers and instances is required. A large number of instances require enormous system resources to access the matrix data, and considerable RBF centers may cause heavy computational burden. Therefore, fitting RBF Kernel to 3D scattered data has not been regarded as computationally feasible for lar