一种利用零交叉点理论的改进SIFT特征提取算法  

An Improved SIFT Operator Based on the Theory of Zero-crossing on Feature Extraction

作　　者：万雪[1] 张祖勋[1] 柯涛[1] 

机构地区：[1]武汉大学遥感信息工程学院,武汉市珞喻路129号430079

出　　处：《武汉大学学报（信息科学版）》2013年第3期270-273,286,共5页Geomatics and Information Science of Wuhan University

基　　金：国家科技支撑计划资助项目(2011BAH12B03)

摘　　要：提出了一种利用零交叉点特征提取的改进SIFT算子用于遥感影像的自动匹配。将图像几何特征引入到尺度空间探测中,获得了重复性更高、更稳定的特征。采用近景数码大旋角数码立体像对和低空航摄立体像对进行了算法测试。实验表明,改进后的SIFT算子应用于遥感影像自动匹配,在特征提取重复率、匹配正确点数、匹配正确率上均有明显提升。An improved SIFT operator based on the theory of zero-crossing is proposed for automatic remote sensing image matching. Firstly, in the improved SIFT algorithm based on zero-crossing theory, the array of scale space pyramid zero-crossing point detection is constructed. Then the search window around the center pixel for the zero-crossing point in the adjacent three layers from the array of scale space detection pyramids is established. If the result of zero-crossing detection is greater than the detection threshold, the center pixel will be marked as a feature point. Feature extraction based on the theory of zero-crossing in Gaussian scale space introduced the image geometry feature in scale space detection, resulting in higher repeatability and more stable characteristics. The experiments carried with close-range digital stereo image which has large rotation angle and low altitude aerial stereoimage showed significant increase in the aspects of feature repeatability rate, correct number of corresponding points and matching correct rate with the improved SIFT operator applied to automatic remote sensing matching image.

关 键 词：零交叉点理论 SIFT算子 特征提取 遥感影像匹配 几何特征 

分 类 号：P231.5[天文地球—摄影测量与遥感]