基于多尺度Harris角点检测的图像配准算法  

An Image Registration Algorithm Based on Multi-scale Harris Corner Detection

作　　者：尚明姝 王克朝 SHANG Mingshu;WANG Kechao(School of Information Engineering,Harbin Institute,Harbin 150000,China;School of Computer Science and Technology,Harbin Institute of Technology,Harbin 150000,China)

机构地区：[1]哈尔滨学院信息工程学院,哈尔滨150000 [2]哈尔滨工业大学计算机科学与技术学院,哈尔滨150000

出　　处：《电光与控制》2024年第1期28-32,共5页Electronics Optics & Control

基　　金：国家自然科学基金(61977020);黑龙江省哲学社会科学基金(21KGB083,22KGB142)。

摘　　要：针对现有多尺度Harris算子算法较复杂、运算量大、精确性一般的问题,提出一种高效简便算法。首先建立多尺度空间,令Harris算子在尺度空间提取特征点,用简化的32维SIFT特征向量描述特征。利用最近邻法匹配特征点;然后采用改进的相似三角形法筛选匹配点,再使用改进的K-means算法对特征点分组,使组内特征点聚集,组间特征点远离;最后应用改进的RANSAC算法在不同组中选取特征点求变换矩阵,避免了选取的特征点距离过近,算法陷入局部最优。实验验证了所提算法的性能。The existing multi-scale Harris operator has a complex algorithmlarge computation amount and low accuracy.To solve the problemsan efficient and simple algorithm is proposed.FirstlyGaussian kernel function is used to build a multi-scale space for imagesand then Harris operator is used to detect feature points in the scale space.The simplified 32-dimension SIFT feature vector is utilized to characterize the feature points.Thenthe nearest neighbor method is used for feature matching,and the modified similar triangles method is used to screen the matching points.The improved K-means algorithm is used to group the feature pointsso that the feature points within the same group are clustered and the feature points belonging to different groups are far apart.Finallythe modified RANSAC algorithm is used to calculate the transform matrix between the two images.The feature points from different groups are selectedso as to avoid the selected feature points being too close to each other and the algorithm falling into local optimum.The experiments verify the algorithm's performance.

关 键 词：图像配准 尺度空间 HARRIS K-MEANS RANSAC 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]