一個由陰影圖像還原物體形狀的演算法

Abstract

在這篇論文中我們提出了一個由灰階陰影圖還原物體形狀的演算法。首先 ，我們針對存在於線性近似影像光度方程式(linear approximation of the imageirradiance equation)中，無法確定此曲面是凸起或凹陷的缺 點，對輸入的陰影圖像作前置處理。在此一步驟中，為了確定此物體的曲 面是凸起或凹陷，我們將二個不同方向的光源投射在此一物體上。同時也 可知道在此區域的局部最高或最低點。由此點，我們試著去找出在每一不 同方向的局部最小灰階值，並改變在此點之後的像素灰階值直到出現另一 局部最小灰階值。此時我們便得到一張和原來不同的灰階陰影圖。然後使 用線性近似影像光度方程式，求得在經過前置處理後之圖像中物體的高度 。物體表面的梯度變化(surface gradients)可由此高度去求得，然後我 們使用亮度限制條件來修正已求得的物體表面梯度變化。亮度限制條件的 功能是在比較原圖和重建後的陰影圖中每一像素之灰階值，據此修正重建 的物體表面梯度變化。最後，由修正後的物體表面梯度變化重建此物體的 高度。 In this thesis, an algorithm in reconstructing shape fromgray- level shading images is introduced. Firstly, the preprocessing approach is applied in the input shading images for the convex or concave ambiguities existed in the linear approximation of the image irradiance equation.In thepreprocessing approach, we use two different illuminantion onto the objet in order to determine that the object is a convex or concave shape. The local maximum or minimum point is also determined. We then try to find the gray-level local minimum pixel in each direction from this local maximum or minimum point.After the gray-level local minimum pixel, each gray-level value is changed inthis direction. Thus, a new gray-level shading image is obtained. The linear approximation of the image irradiance equation is used here to derive the surface height of the new gray-level shading image. The surface gradients definition is then used to obtain the surface gradients from this surface height. We then use the surface gradients modification with brightness constraint to obtain the correct surface gradients. The function of the brightness constraintis to compare the gray-level value in the original shadingimage with the recoveredgray-level valre derived from the reconstructed shape. Finally, the depth mapis reconstructed from the surface gradients.