基于兴趣点的遥感影像可匹配性度量

Matching probability metric for remote sensing image based on interest points

为了提高基准图制备的有效性,研究了遥感影像的可匹配性,并针对点特征提出了基于兴趣点度量遥感影像可匹配性的度量指标。首先,利用非下采样Contourlet变换的高频系数提取图像的局部极值点作为兴趣点;然后,利用兴趣点的高频系数和方向特性定义兴趣点的幅值特性和结构特性。最后,分析兴趣点幅值特性和结构特性与实际匹配概率的关系,构建可匹配性度量指标。实验结果表明,可匹配性度量指标与实际匹配概率的线性相关性高于0.9,Spearman相关系数高于0.85,评估结果准确性高、单调性好。利用该指标对图像可匹配区域进行筛选,其制备的基准图的平均匹配概率大于95%,比传统方法提高了15.4%,改善了基准图制备的效率和可靠性。

To improve the effectiveness of reference map preparation,the matching probability of a remote sensing image was investigated.A matching probability metric for the remote sensing image was proposed based on interest points.Firstly,the local extreme points obtained from high frequency coefficients of non-subsampled Contourlet transform were set as interest points.Then,the amplitude and structure properties of interest points were defined by the frequency coefficients and direction information.Finally,the matching metric was established based on the analysis of the relationship between amplitude and structure properties and real matching probability.Experiments show that the correlation coefficient of the metric and real matching probability is greater than 0.9 and the Spearman rank order correlation coefficient is greater than 0.85 with high accuracy and consistency.Matching region extraction by using the proposed index gives the average matching probability of more than 95％ for reference maps,increasing 15.4％ more than traditional methods.It improves the efficiency and the reliability of reference map preparation.