摘要
针对尺度不变特征变换(SIFT)算法的匹配结果存在错误匹配点以及冗余匹配点的问题,提出一种改进的图像匹配算法。该算法将SIFT算子的尺度空间极值点作为初始特征点,用Harris角点检测算子对初始特征点进行筛选,选择高对比度的点作为最终特征点。接着设置合适的欧氏距离阈值进行粗匹配。由于SIFT得到的匹配点集存在冗余与错误匹配,改进的算法在匹配后再进行一次逆向匹配,最后,利用RANSAC算法进行纠错,得到正确的匹配特征点对。实验结果表明,在图像有旋转、平移、光照等情况下,该算法稳定、可靠。
Puts forwards an improved image matching algorithm aiming at the problems of that Scale Invariant Feature Transform algorithm has a few false and repeated matching feature points. The improved image matching method takes the extremum got by the SIFT descriptor as origi-nal keypoints, then filters the original keypoints by Harris corner detection operator. Selects points which have a high contrast as final points and match final points by using the proportion of the Euclidean distance. As there are some redundant and error matching points, conversely matches both images based on SIFT matching. Finally uses RANSAC algorithm to select the correct match keypoints. The re-sults show that the algorithm is stable and reliable under the change of rotation, translation, light, contrast and so on.
