面向边缘特征的遥感影像可匹配性度量

Remote Sensing Image Matching Performance Measurement for Edge Feature

为了提高基准图制备的有效性,需要对前期遥感影像的可匹配性进行预测.在边缘特征中,提出基于块逆概率差纹理基元共生矩阵的遥感影像可匹配性度量方法.首先,利用边缘密度将遥感影像分为不可匹配区和潜在可匹配区,并从潜在可匹配区选取部分图像样本作为训练集;其次,对样本图像构建块逆概率差纹理基元共生矩阵,计算边缘特征向量;再次,利用仿真实验计算样本图像实际匹配概率,采用支持向量机回归方式利用边缘特征向量构建匹配概率预测模型;最后,对整幅遥感影像采用匹配概率预测模型预测匹配概率.实验证明,该方法预测的匹配概率与实际匹配概率平均平方误差低,平方相关系数强,并且对于灰度校正后的同一卫星遥感影像其预测模型是通用的,不仅能满足遥感影像可匹配性度量需求,并能对匹配算法选择提供决策支持.

In order to improve the effectiveness of reference map production, it is necessary to predict matching performance for early remote sensing images. For edge feature, the saliency and stability of edge determines matching performance. An algorithm for remote sensing image matching performance was proposed based on block difference of inverse probabilities and texture cell cooccurrence matrix. Firstly, remote sensing image was divided into potential matching regions and no matching regions based on edge density, and training images were extracted from potential matching regions. Secondly, edge feature vector was computed by block difference of inverse probabilities and texture cell cooccurrence matrix （BDIP-TC-CM）. Thirdly, on basis of the real matching probability computed by simulation experiment, matching probability predicting model was built by support vector regression based on edge feature vector. Lastly, matching probability was predicted for the whole remote sensing image based on the matching probability predicting model. The experimental esult shows that the mean squared error between the predicted matching probability and real matching probability is small and the squared correlation coefficient is high. The model is general for the same satellite images after gray level correction. It can meet the demand of remote sensing image matching performance measure, and provide decision support for selecting matching algorithm.