Please use this identifier to cite or link to this item:
標題: 高斯與平均曲率在形貌影像切割之研究
Study of Gaussian and Mean Curvatures in Range Image Segmentation
作者: 張哲仁
關鍵字: range image;形貌影像;initial segmentation;region growing;Gaussian curvature;mean curvature;variable-order surface fitting;影像切割;初略切割;區域成長;可變階數表面擬合
出版社: 機械工程學系
電腦視覺系統從3D世界中所取得龐大的影像資訊,要如何轉換成有用的資訊。若直接藉由電腦來闡釋原始影像資訊是有困難的,通常最適宜的方式先做影像切割(Image segmentation)的步驟,將影像分割成一件件的低階元件來做後續的處理,所謂低階元件就是指藉由具有相同特性的像素群體組成的。而本論文目的就是建立一套3D形貌影像切割演算法,切割演算法大致分為初略切割與區域成長兩大部分。首先,使用表面曲率符號標記將形貌影像做初略的分割,接著做區域生成,成長方法是採用可變階數表面擬合重新定義表面型態,以不同階數做曲面擬合,直到預設的誤差範圍內。文中之實驗部分所用到的輸入影像是用電腦模擬形貌影像的資料,而影像中元件係以平面、圓柱面及圓錐面等幾何形狀所組成,切割的效果以6張影像來呈現整個演算法的過程及結果。

In this paper we present a region based segmentation algorithm for range image . The algorithm can be viewed as a two-stage process: (1)initial segmentation,(2)region growing . First, we segment surface into a set of regions in the range image by Gaussian and mean curvature signs, and then produce a surface type label image . Second , we perform an iterative region growing using variable-order surface fitting . The fitting input is a seed region from surface type label image . In order to determine the seed region , we isolate the largest connected region of any surface type , and then shrink the region until appropriate size is obtained . Using the seed region begins to grow iteratively and then refine the surface definition . Experimental results show the algorithm performance on six range images.
Appears in Collections:機械工程學系所

Show full item record
TAIR Related Article

Google ScholarTM


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.