采用多尺度滤波的遥感图像关键点检测方法  

Key-points feature detection using multi-scale filtering for remote sensing image

作　　者：孙建中[1,2] 熊忠阳[1] 张玉芳[1] 

机构地区：[1]重庆大学计算机学院,重庆400030 [2]商丘师范学院,河南商丘476000

出　　处：《计算机工程与设计》2012年第8期3125-3129,共5页Computer Engineering and Design

基　　金：中央高校基本科研业务基金项目(CDJXS10180004);中国博士后科学基金项目(20070420711);重庆市自然科学基金项目(CSTC2008BB2191)

摘　　要：针对关键点检测过程中遥感图像会因为局部结构变形发生误检测的问题,提出一种鲁棒的关键点检测新方法。该方法首先基于映射策略和提升结构实现非下采样多尺度滤波器,用于获得平滑图像;然后根据图像尺度空间理论和尺度不变特征变换生成高斯差分尺度空间,并利用有限差分计算尺度空间采样点局部极值,子像元插值和Hessian矩阵删除不稳定极值点;最后用梯度分布直方图描述关键点。在仿真实验平台上,与现阶段方法的关键点计算时间和图像匹配性能做比较分析,实验结果表明,改进的方法有较低的计算时间复杂度,具有较好的鲁棒稳定性。Focusing on the increasing of the incorrect remote sensing image key-points caused by deformation of local structure, a novel approach for remote sensing image key-points detecting is proposed. At first, a filter is designed using mapping approach and lifting structure for obtaining smoothing image. Then, the scale-space difference-of- Gaussian is produced based on scalespace theory and SIFT. The local extremes of sample points are calculated using finite difference in scale space, followed by sub- pixel interpolation and Hessian matrix to reject unstable extremes, and finally key-point descriptor is represented through orientation histograms. The time complexity and matching performance of the proposed methods are analyzed comparatively by experiment. Experimental results suggested that this algorithm reduces the time complexity and improve the image-matching robust stability.

关 键 词：遥感图像 关键点 多尺度滤波 尺度不变特征变换 图像匹配 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]