尺度不变特征点提取  

Scale Invariant Feature Points Detection

作　　者：刘立[1,2] 谭邦[1] 肖建田[1] 

机构地区：[1]南华大学计算机科学与技术学院,湖南衡阳421001 [2]华中科技大学电信系,湖北武汉430076

出　　处：《南华大学学报（自然科学版）》2009年第1期85-88,共4页Journal of University of South China：Science and Technology

基　　金：国家自然科学基金资助项目(60475024);航天技术创新基金资助项目(2006AA09Z203)

摘　　要：结合传统的角点检测算法,提出一种基于高斯金字塔匹配的尺度不变特征点提取算法PBSI,首先建立高斯金字塔,在每层图像中检测Harris角点,根据高斯尺度理论自上而下找到它们在不同层的对应点,所有层上都有的匹配点就是最终的尺度不变特征点.实验结果表明本文算法具有计算简单、抗噪声能力强、稳定性好的优点.Combined with traditional comer detection algorithm, a scale -invariant feature extraction algorithm is presented based on Gaussian pyramid matching method （PBSI）. After building Gaussian pyramid, the Harris comers is detected in each level. In order to find the counterparts of different level a pyramid matching method is taken top - down accord- ing to theory of Gaussian scale. The comers can be found in each level is the final feature points. The results show that the algorithm has low computation cost, strong ,anti - noise capability and good stability.

关 键 词：金字塔匹配 尺度不变 HARRIS角点 高斯尺度 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]