基于数字视差模型的立体影像最小二乘密集匹配

提出一种基于数字视差模型的立体影像最小二乘密集匹配方法。在高精度初始匹配点的基础上构建数字视差模型,利用多项式内插的方法预测出加密点的初始同名点位置,以结构相似性指数作为相似性测度进行加密点的粗匹配,然后利用最小二乘匹配算法将匹配精度提高至子像素级,最后由随机抽样一致性方法剔除误匹配点。实验结果表明,该方法能较为迅速地预测出加密点的同名点初始位置,且具有较高的匹配正确率,能得到十分密集的匹配点集,可以高效地应用于立体影像的密集匹配任务中。

Least Square Dense Matching of Stereo Images Based on the Digital Parallax Model

A least square dense matching method based on the Digital Parallax Model has been proposed.The DPM is constructed under the guideline of some high accurate matching points,initial corresponding points of pass points are predicted by polynomial interpolation,with the structure similarity taken as similarity measure,coarse matching is executed.Then,the degree of matching accuracy is promoted to sub pixel through least square matching strategy,At last,some error matching points are eliminated after the utilization of random sample consensus.Experimental results suggest that the initial corresponding points of pass points can be anticipated quickly and the correct rate of matching is pretty high owing to the proposed method which can be used in dense matching mission of stereo images efficiently.