三維影像對齊與紋理顏色校正

Abstract

在電腦視覺裡，如何呈現一個完整的模型和高畫素的彩色紋理是一個重要的議題。在 本論文中，我們主要著重於影像對齊及該模型之紋理顏色校正。其中，影像對齊方面， 使用的是最經典的且最有效率的剛體對齊演算法-最近點迭代法，來解決來自不同三維 影像攝影機的資料，將他們完整接合成一個模型，並應用該演算法來解決由於規格不 同造成尺寸不對的問題。紋理顏色校正方面，我們根據兩個影像間重疊部份，來將顏 色進行修正，使得兩個影像顏色差異減少，並對兩影像間的接合處做調整，將可能還 存在的接縫平滑化，讓整個模型看起來更為真實且完美。

In computer vision, presenting a complete model and high quality texture is an important issue. In this thesis, we focus on 3D image registration and color calibration of the texture. In the image registration part, we use Iterative Closest Point, the most classical and efficient rigid registration algorithm, to combine data coming from different sensors into a complete model. We also apply this algorithm to solve the data size mismatch problem caused by specification differences between the sensors. In the texture color calibration part, we calibrate colors based on the overlapping region between two images to reduce the difference of their colors. This step reduces sudden change of color at the boundary between images and makes the model more realistic and perfect.