一种基于参考点距离的SIFT特征点匹配算法

摘要

To address the high time cost of feature point matching in scale invariant feature transform ( SIFT ) , a new SIFT feature point matching algorithm based on the distance to reference point—DRP algorithm is put forward. Firstly, distances from the reference point to every feature point to be matched is computed using DRP algorithm. Then, these distances computed previously is ordered and saved in a dataset named as distance of ordering. Next, distances from the reference point to the feature point to be queried is also computed. After that, the nearest neigh⁃bor of the distance in distance of ordering is retrieved with binary search and returned as index of center. Finally, the nearest neighbor of feature point to be queried is searched one by one in a certain range whose center is index of center. It is proven by experiment tested on ACF ( affine covariant features) pictures from VGG( visual geometry group) laboratory that DRP algorithm can effectively decrease the time cost of SIFT feature points matching without loss of matching results compared with the classical SIFT algorithm.%针对SIFT特征点匹配时间消耗大的问题，提出了一种基于参考点距离的SIFT特征点匹配算法—DRP算法。该算法首先计算一次所有待匹配特征点到参考点之间的距离，对之进行快速排序并保存。然后计算待查询特征点到参考点的距离，并在已排序的距离中使用二分法搜索返回此距离的最近邻。最后以此最近邻为中心，在有限范围内搜索待查询特征点的近似最近邻。 VGG实验室ACF图片库的测试结果表明，相比于经典的SIFT算法，DRP算法可以在不损失匹配效果的前提下，有效降低SIFT特征点匹配的时间消耗。