SIFT算法中融合全局不变矩的特征变换  

Feature Transform of Global Invariant Moments in SIFT Algorithm

作　　者：张磊[1] 胡燕[1] 

机构地区：[1]武汉理工大学计算机科学与技术学院,湖北武汉430070

出　　处：《微电子学与计算机》2017年第6期67-71,共5页Microelectronics & Computer

摘　　要：提出了一种融合全局几何特征的SIFT描述算子.该算法以特征点为中心,进行对数极坐标变换并建立同心圆区域,通过分析HU矩在离散状态下的变换情况,提取出在离散状态下具有RSTC不变性的不变量作为同心圆区域的特征向量.采用欧式距离作为度量标准进行图像匹配实验对比.结果表明,此算法通过结合全局特征与局部特征,提高了图像在旋转与尺度变化、视点变化、光照变化和图像模糊等各种变化下匹配的鲁棒性.This paper present a SIFT descriptor that combines global geometric in this paper. The algorithm takes the feature point as the center, carries out the log polar coordinate transformation and establishes the concentric circle region, By analyzing the transformation of HU moments in discrete state, extracting the Feature vector of the invariant of the concentric circles in the discrete state with RSTC invariance. Tacking the Euclidean distance as a metric for image matching experiments. Experiments show that the algorithm which combines the global with local feature improves the robustness of image matching under various changes such as rotation and scale changes, viewpoint changes, illumination changes and image blur.

关 键 词：尺度不变化特征 对数极坐标变换 几何特征 不变量 

分 类 号：TP31[自动化与计算机技术—计算机软件与理论]