结合图像信息和几何关系的配准算法

Past precise registration algorithm combining image information and geometric relations

对图像配准中的特征点检测方法和描述方法进行了研究,提出了一种快速准确的图像配准算法.使用尺度空间理论进行特征点的检测,通过两种不同的特征点描述子对特征点进行描述:一种是基于图像信息的描述子,该描述子使用尺度空间理论上的尺度不变Harris算子的自相关矩阵来描述特征点;第二种是基于空间上的几何关系的描述子,提出了一种融合局部结构特征和全局信息的描述子.在图像配准的目标函数里,将特征点匹配矩阵和图像变换矩阵分解,分别使用两种不同的描述子对二者进行迭代求解.该算法结合了两种特征点描述方法的优点,实验结果表明该方法快速、准确,具有配准精度高和计算时间短的优点.

A fast and precise algorithm on image registration was proposed through the research on the methods of feature point detection and description .First ,the feature points were detected by a filter designed based on the scale space theory .Then ,two descriptors of feature point were used for point description .One is the descriptor based on the image information ,with the autocorrelation matrix of the scale invariant Harris operator to describe points .The other is the descriptor based on the space geometric relations among points ,w hich integrates the local structure and global information of ima-ges .The two components in the objective function of the image registration ,the matching matrix and the image transform matrix were solved by the two descriptors iteratively .By combining with the ad-vantages of the two description methods ,the algorithm achieves high alignment accuracy and small computational volume .