基于Contourlet旋转不变特征的遥感图像检索

A New Image Retrieval Method Using Contourlet-based Rotation-invariant Feature for Remote Sensing Images

除了具有纹理属性外,遥感图像中还包含大量的结构性边缘特征,如何有效捕捉这些特征中的信息进行检索,成为提高遥感图像检索效率的关键。文章依据Contourlet在离散域的多尺度几何分析的功能,提出一种利用Contourlet子带能量分布特性提取原始图像在多个尺度下的边缘方向信息进行检索的方法。针对Contourlet变换捕获不同特征方向能力的差别,文章采用正交补偿法加以改正,并通过傅里叶算子的处理,实现基于结构性边缘特征旋转不变的图像检索。实验结果表明,该检索算法对含有丰富规则边缘特征的遥感图像更为有效。

There are plenty of structure edge features in remote sensing images besides texture features. How to capture the information contained in these features is a key point to retrieve remote sensing image effectively. Given that Contourlet transform has the function of multi-scale geometric analysis in the discrete domain, we have put forward a novel approach for image retrieval, which extracts the energy features of Contourlet subbands to obtain the edge direction information under different scales. For the ability differences in capturing different direction features, we have adopted an orthogonal compensation method to remedy them, and realized rotation invariant image retrieval through processing them with Fourier operator. Experiments on remote sensing images demonstrate this retrieval algorithm is especiauy effective for those images containing rich structural edge features.