基于Harris算子和方向场的图像配准算法  

Image registration algorithm based on Harris operator and direction field

作　　者：曾新贵 陶卫[1] 颜发才 赵辉[1] 

机构地区：[1]上海交通大学电子信息与电气工程学院,上海200240

出　　处：《计算机应用》2016年第A01期146-148,共3页journal of Computer Applications

摘　　要：为解决光照剧烈变化情况下图像难以精确配准的问题,结合Harris角点提取和方向场算法方向提取特征稳定的优势,提出了一种对光照剧烈变化不敏感的图像配准算法。首先使用Harris算子提取特征点,接下来使用方向场算法计算特征点及其邻域像素点的方向特征,以提高特征点的抗光照变化特性,并构建特征点的描述向量。再使用双向最近邻算法匹配特征点,根据相对偏移量计算,剔除误配点,最后使用随机抽样一致(RANSAC)算法提高特征点匹配精度,得到图像间的几何变换参数,完成图像配准。实验表明,该算法可以实现像素级精度的图像配准。相比尺度不变特征变换(SIFT)算法,该算法可以提取更多的特征点,配准成功率提高22%,配准精度提高2.34%,在光照剧烈变化且在SIFT算法无法配准的情况下依然能准确配准图像,具有更好的抗光照变化特性。A new image registration algorithm was proposed to solve the problem that images were difficult to be accurately resistered when the light intensity changed,which combined the advantages of the Harris corner detection and direction field.First using the Harris operator to extract the feature points,and the direction field algorithm was used to compute the directional characteristics of the feature points and their neighbors,which could improve the anti-illumination change characteristics of the feature points,then constructing the feature description.The two-way nearest neighbor algorithm was used to match the feature points and eliminate the false matching points according to the relative offset calculation.Finally RANdom Sample Consensus（ RANSAC） algorithm was used to improve the matching accuracy,calculate image transform parameters to accomplish the image registration process.Experiments show that the algorithm can achieve pixel level accuracy of image registration and extract more feature points than Scale-Invariant Feature Transform（ SIFT） algorithm.The registration success rate and the registration accuracy are improved by 22% and 2.34% respectively compared with the SIFT algorithm,and has better anti-illumination change characteristics.

关 键 词：图像配准 HARRIS算子 方向场 尺度不变特征变换 随机抽样一致性 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]