SIFT算法的改进

【目的】为了减小三维重建的重投影误差,提出一种改进的SIFT(Scale Invariant Feature Transform)算法。【方法】首先使用SIFT提取和匹配特征点,将这些匹配点作为归一化互相关(Normalized Cross-correlation,NCC)的初始匹配对;然后使用特征点的主方向对局部图像进行旋转校正;最后计算该初始匹配对NCC系数并将相似地貌中的误配点剔除。【结果】该方法剔除了大量的误配点,提高了特征点的正确匹配率和重建结果的精度。【结论】改进的SIFT算法能够得到更为准确的匹配点对,获得较好的重建效果。

The Improvement of SIFT Algorithm

Objective]In order to reduce the reprojection error of reconstruction, an improved SIFT algorithm is proposed. Method]Firstly, SIFT is used to detect and match the features. These match points are used as the initial match on normalized cross-correlation （ NCC ） . Then dominant direction of feature points is used for rotation correction of local image. Finally, the coefficient of normalized cross-correlation matching（ NCC） is calculated and the mismatches points in the similar geographical environment are removed.Result]This method removes a lot of mismatches points, and improves the rate of correct matching and precision of reconstruction.Conclusion]Experiment results show that the improved algorithm can achieve reconstruction effect.