三維掃描使用日漸廣泛,如何建造一個趨近實體的三維模型為重要議題之一。透過掃描後將會得到模型的三維深度資訊及色彩資訊,而本篇著重於色彩資訊的校正。有別於大部分建立三維模型時給予單點顏色的形式,我們利用紋理對應(Texture mapping)以三角網格為單位賦予實際影像的色彩以便提高模型的解析度,並用影像對齊(Image registration)的方法減少影像重疊的錯位(misalignment)。對於影像重疊的部分進行顏色修正,減少兩影像接合處的差異。最後,以影像混合(Image blending)的方式得到色調較為一致的影像。
In recent year, three-dimension scanner has widely used, so it is an important issue to approach a realistic object. The depth information and the color information of the object are obtained by the scanner. In this thesis, we focus on the calibration of the color information. Rather than presenting a model that is colored with simple per-vertex color, we use the texture mapping to wrap model with realistic images and increase the resolution. Moreover, we use image registration to reduce the misalignment between adjacent patches. For the overlapping region, there is a color correction method to make image stitching seamless. Finally, image blending makes the color tone of images consistent.