为解决尺度不变特征变换( SIFT)算法在图像发生旋转和尺度变化时产生的错误匹配问题, 提出一种新的算法。根据 SIFT提取的关键点信息, 利用正确匹配点对间的旋转不变因子和尺度不变因子来剔除 SIFT误匹配点, 然后对保留下来的特征点进行聚类分析, 对目标图像进行识别判断, 并通过实验将该算法与双向匹配算法和随机抽样一致性算法( RANSAC)进行比较。实验结果表明, 该算法能够有效地剔除误匹配点, 且误剔除率低。剔除误匹配点后再进行图像检索, 图像的漏检率和误检率都大大地降低了。

A new algorithm was put forward to solve SIFT mismatching problem that was caused by the image rotation and scale changes. It used the image rotation-invariant and scale-invariant factors of the right matching points to eliminate the mismatching points with the key information of SIFT extraction. Then it analyzed the reserved feature points by clustering to recognize the target images and was compared with the bidirectional matching algorithm and RANSAC algorithm through the experiments. The experiment results showed that this algorithm can effectively eliminate the false matching points and the false rejection rate is low. Images are retrieved after the mismatching points excluded, and the miss rate and the false detection rate of the images are greatly reduced.