摘要
针对SIFT算法存在内存消耗多、运算速度慢的问题,采用金字塔和分块策略,首先对原始影像进行粗配准,然后再作分块影像匹配,在匹配过程中根据局部熵过滤掉冗余特征点,并使其均匀分布于影像,以实现精确配准。对错误匹配点先利用相关系数初步剔除错误点,再利用极线约束对错误匹配点进行精细剔除,最后将RANSAC算法应用于剩下的匹配点,进一步提高匹配精确度。
Regarding to the issues of large memory consumption and slow computing speed in SIFT algorithm, this article used pyramid and partition strategy to coarse register the original images at first. And then, the matching of block images was adopted to achieve accurate registration. During the processing of matching, redundant feature points were filtered out and uniformly distributed on the image based on local entropy. Concerning the error matching problem in SIFT algorithm, the paper eliminated error points by the correlation coefficient initially, and then used epipolar constraint to accurately eliminate the error points. At last, remaining points were processed by RANSAC algorithm to further improve the accuracy of matching.
出处
《地理空间信息》
2017年第2期69-71,75,共4页
Geospatial Information
关键词
SIFT
影像配准
极线约束
点特征
SIFT
image registration
epipolar constraint
point feature