一种提高SIFT特征匹配正确率的方法

针对尺度不变特征变换(SIFT)特征匹配算法存在计算量大、实时性差、误匹率高的问题,提出一种基于距离比率准则的方法来去除SIFT特征匹配中的错误匹配。传统的方法是采用随机选取一致性(RANSAC)方法选取出正确的匹配对,但是需要通过反复迭代,复杂、耗时并且仍存有部分误匹配的现象。改进后的方法直接通过两条匹配直线斜率的一致性判断,剔除不在斜率范围内的匹配,此方法算法简单,省时高效,从而较大提高了特征匹配的正确率。实验结果表明,通过采用距离比率准则方法具有较高的匹配精度,同时减少了匹配的时间,使实时性得到提高。

A method for improving matching accuracy of SIFT features

Aiming at solving the problems of large calculating scale, poor real-time performance and high false-matching rate in Scale Invariant Feature Transform (SIFT) feature matching algorithm, this paper presents an improved SIFT feature matching algorithm based on distance-ratio criterion to eliminate the matching errors. Traditional random selection consistency (RANSAC) method is used to select the correct matching pairs. However, it is complex and time-consuming for the repeat iteration process as well as its occasionally matching error occurrences. In the proposed improved algorithm, slope consistency of the two matching lines is directly compared while the matching pairs beyond the slope range are eliminated. This improved algorithm is simple and efficient. Experiments show that this improved algorism based on based on distance-ratio has great matching accuracy, saving matching time with performance and efficiency enhanced.