影像透明度分析於結構光影像解碼之應用與3D影像重建

Abstract

在影像處理中，由影像中擷取其中的主體，或稱前景為一個重要的課題。藉由譜方法影像擷取，我們分析影像的透明度稱為圖層透明度。藉由此方法，圖層透明度可導出一個二次非線性方程。為了計算此二次非線性方程，在我們的論文中，藉由使用者給予的三元圖（Trimap），我們提出了與以往不同的演算法。在有限制條件的情況下，我們利用有限元素法計算此二次非線性方程來提高計算的準確度及減少計算量。除此，根據譜方法影像擷取，我們將此技術應用於三維影像重建。我們使用了結構光的技術並結合圖層透明度的結果，達到編碼的目的，藉此重建三維影像。

From the computer vision, it is important to extract a foreground object from an image. With the Matting Laplacian, we compute the foreground opacity, called alpha matte. It can be derived as quadratic cost function in alpha estimation and we propose to solve the cost function. Our method is that a trimap can be given by user, and we employ the Finite Element Method combined with constrained quadratic programming. Under the spectral matting concept, we apply this technique to 3D image modeling. In order to construct the 3D surface from 2D images’ information, various structured light pattern are used to encode the correspondence between projector and camera. As a result, 3D geometric information can be computed by elementary projective geometric.