從大型三維影像中擷取適應式等值亮度表面

Abstract

從磁振造影或斷層掃描之影像當中擷取的等值亮度表面，可以幫助我們了解潛藏在影像中實體的三維結構。利用走訪立方體法可以擷取單一解析度的三角形網格，但針對非常大型的三維影像可能會產生過多的三角形，造成儲存及顯像上的困難。針對這個問題，我們提出一個適應式等值亮度表面擷取法，可以在實體表面變化較劇烈的地方用較精細的三角形來建構，而表面較平坦之處則用較大的三角形來建構，如此可以兼顧三角形數量與表面模型準確性。首先，我們會在三維影像當中，以多維搜尋樹的方式快速找到有等值亮度表面通過的區域，然後利用一個快速演算法來分析其表面變化情形，並產生適應於表面的多重解析度三角形網格。根據我們的實驗結果，經由我們所提出的方法所產生的網格，其三角形個數最多只有走訪立方體法的四分之一，且並沒有損失過多的表面模型精準度。

Isosurface extracted from volume data, such as CT, SPECT and MRI, can provide 3D structure information. The marching cubes algorithm has been extensively applied in isosurface extraction from volume data. This algorithm generates spatially regular triangle meshes and can be very storage-consuming and hard rendering when the volume data is huge. We present an adaptive isosurface extraction algorithm which can reduce the triangle number of isosurface by adapting to the variation of the isosurface. The adaptive isosurface contains larger but fewer triangles for low variation area and smaller but more triangles for high variation area. We use kd-tree method to rapidly locate the area containing the isosurface. Then a novel recursive algorithm is applied to analyze the volume data and to generate the triangles with sizes adaptive to the surface. According to our experiments, the triangle number obtained by using the proposed method can be at least four times fewer than that can be obtained by using the marching cubes method without losing significant surface accuracy.