基于RANSAC的SIFT匹配阈值自适应估计

Estimate Threshold of SIFT Matching Adaptively Based on RANSAC

针对基于欧氏距离比值作为图像尺度不变特征变换(SIFT)特征匹配相似性度量时,距离比阈值难以设置最优,且固定距离比阈值易引起误匹配或漏匹配等问题,引入随机抽样一致性(RANSAC)算法。该算法对SIFT匹配算法中的距离比阈值进行自适应优化,确定最佳的阈值,再利用双向匹配的方法剔除误匹配点。实验结果表明,针对不同的实验图像,所提算法都能自适应地求解出一个最优的比例阈值,使得匹配点数最多,同时具有较高的匹配正确率,经过双向匹配的策略优化后效果更好。

When matching images with scale invariant feature transform(SIFT),the Euclidean distance between feature vectors is used as the similarity measurement.But it was difficult to get the best distance ratio.Moreover,when the ratio was a constant,there would be some problems of error matching or matching leakage.Deal with the problem,the Random Sample Consensus(RANSAC)algorithm was introduced.Optimize the ratio in the process adaptively,and we can get the best threshold.SIFT-based image matching algorithm was analyzed,and a bi-direction matching was used to improve the accuracy of image matching and ensure the correctness of matching at maximum level.Finally,the experiment results show that the proposed methods can obtain an optimal threshold for different images.It can get the most matching points and a better matching rate,and by bi-direction matching,better results can be got.