基于SIFT的自动匹配策略

摘要

For the purpose of actualizing automatic image matching, a new Scale Invariant Feature Transform (SIFT) matching strategy is presented. Firstly, compared with the given basal ratio of Euclidean distance from the closest neighbor to the distance of the second closest of features, the coarse matching collection of SIFT is obtained. Then, obtain the histogram of coarse matching corresponding points' main direction angle differences and the histogram of minimum Euclidean distances. The precise matching is taken under the conditions that the main direction angle differences fall into the neighborhood of histogram's peak value. And the minimum Euclidean distance is smaller than the corresponding distance of histogram's main peak. Finally, obtain the histogram of precise matching corresponding points' ratio. The false matching is eliminated under the conditions that the ratio of Euclidean distance from the closest neighbor to the distance of the second closest is smaller than the corresponding ratio of histogram's main peak. The experimental results prove that the proposed method is stable and reliable even if the image has some view point, illumination, rotation and scale variation. The proposed method can select the matching threshold automatically without manual intervention.%为了实现图像的自动匹配,提出了一种新的SIFT匹配策略.首先,给定一个基本的欧式距离最小值与次小值的比率值,对SIFT特征对进行判断,得到粗匹配集合.然后,统计粗匹配对的主方向角度差直方图和最小欧式距离直方图,以主方向角度差落入直方图峰值所在邻域和最小欧式距离小于直方图主峰所对应的距离值作为匹配条件进行精匹配.最后,统计精匹配对的比率直方图,以欧式距离最小值与次小值之比小于直方图主峰所对应的比率值作为控制条件,剔除误匹配.实验结果表明:在待匹配图像有一定程度的视点、光照、旋转、比例变化等情形下,该方法具有稳定、可靠的特点.该方法能自动选择匹配阈值,实现无人工干预的自动匹配.