基于尺度空间约束的融合特征点匹配方法  

Matching method of combinative feature based on constraint in scale space

作　　者：林强[1] 李征[1] 吴仲光[1] 

机构地区：[1]四川大学计算机学院,成都610065

出　　处：《四川大学学报（自然科学版）》2015年第6期1249-1254,共6页Journal of Sichuan University(Natural Science Edition)

基　　金：国家自然科学基金(61471250)

摘　　要：本文提出一种多尺度特征匹配的空间约束机制,Combinative Feature based on Constraint in scale space(CFCS-SIFT),该约束机制以SIFT特征点的尺度为基础,对多尺度空间中检测到的DOG特征点与Harris角点提供匹配空间约束,以提高正确匹配点对的数量.基于该约束机制,构造了一种融合DOG特征提取、Harris角点提取原理的SIFT描述符提取与匹配方法,该方法在多尺度空间中提取DOG特征点、Harris角点,并根据特征点的空间、坐标参数获取SIFT描述符.在将DOG特征点和Harris角点相融合并生成SIFT描述符的基础上,设定尺度阈值,根据尺度阈值对检测范围进行空间约束,在约束范围内查找特征点,采用BBF(Best Bin First)算法,并用欧氏距离作为度量函数进行特征点的匹配,最后用RANSAC对匹配点对进行筛选纠错.通过大量实验证明,该算法能够找到更多匹配点对,正确匹配点对相对于不具有空间约束的融合特征点匹配方法增加了15%左右.In order to get more correct feature points, this paper present a new method of image matching, which combining DOG feature with Harris corner, and setting a constraint of detecting the feature points in scale space. The feature points were detected by scale-space extrema of DOG and Harris operator. The Harris corners should be detected in different scales of the image to keep the scale-invariance. Then, the main orientation for each feature point was calculated, and the feature point descriptors were obtained. The feature points should be searched in cycles. The number of cycle was calculated by the scale threshold of feature points, which was given by some specific value. The feature points were found in the constraint areas, and the size of area was calculated by the scale of feature points. Lastly, the feature points were matched by 13BF algorithm, and corrected by RANSAC algorithm. The experiment reaults show that the new method can obtain about 15% more correct matches than the method without searching constraint.

关 键 词：图像匹配 SIFT HARRIS角点 尺度空间约束 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]