该文提出一种基于空间约束的快速鲁棒特征(SURF)匹配优化算法,称为SC-SURF。首先通过SURF算法检测和匹配图像的特征点。然后根据最近邻比例越低其匹配精度越高的特点,得到按最近邻比率排序的匹配点。并以最优匹配点作为参考点生成新的坐标系,利用空间位置关系地图对每对匹配点进行编码。同时为了简化随机抽样一致性(RANSAC)算法,选择尽量少的最优匹配点对作为RANSAC的代表测试数据集,并由该测试数据集拟合目标投影变换矩阵。最后结合匹配点间的空间位置关系和简化的RANSAC算法对匹配点进行几何校验。实验表明该方法在达到良好匹配精度的同时,具有鲁棒性强,匹配速度快的优点。
An optimization algorithm based on Spatial Constraint for Speeded Up Robust Feature (SURF) matching is proposed, called SC-SURF. First, SURF is used for the image feature point detection and matching. Then the matched points are ranked according to the principle that the lower is the ratio of the nearest neighbor, the higher is the matching accuracy. A new coordinate system is created based on the optimal matched points. Every pair of matched points is encoded using the relative spatial map. At the same time, representative data sets are constructed to simplify RANndom SAmple Consensus (RANSAC) by using a minimal number of optimal matches. The target homographic matrix is fitted based on the representative data set. Finally, the spatial verification is performed using the relative spatial map among the weighted matched points and simplified RANSAC. Experiments demonstrate that SC-SURF algorithm achieves good robustness and high speed while maintaining high matching accuracy.