一种用于三维重建的彩色Sift准稠密匹配算法

摘要

针对复杂光照条件下Sift算法对彩色图像匹配能力较差,基于Kubelka-Munk理论,提出了一种适用于未标定图像的准稠密立体匹配算法,有助于更精确地进行三维重建.该算法首先求出彩色图像各个像素的颜色不变量,提取彩色特征点并通过构造彩色Sift特征描述子进行初匹配,采用RANSAC鲁棒算法消除误匹配生成种子点；然后依据视差约束提出一种基于视差梯度均值自适应窗口方法,根据视差梯度均值调整搜索范围；最后采用最优先原则进行区域增长.实验证明,该算法能获得比较满意的匹配效果,是一种有效的用于三维重建的准稠密匹配算法.%This paper proposed a quasi-dense matching algorithm based on the Kubelka-Munk theory for uncalibrated images, it contributed to more accurate 3D reconstruction. Because Sift algorithm was less capable of matching for color images under changing illumination. First, it found color invariant of each pixel in color images for extracting color feature points and constructing color sift feature descriptor to match, using RANSAC robust algorithm to eliminate false matches for generating seed points. Then it presented a adaptive window method to adjust search scope, which was based on the mean of disparity gradient according to the parallax constraint. Finally, it used the principle of the highest priority for regional growth. Experiments show that this algorithm can obtain satisfactory quasi-dense matching results, and it is an effective dense matching algorithm for 3D Reconstruction.