基于卡通纹理分解和NSCT的卫星云图边缘提取

为方便提取云图的缓变边缘,提出了基于卡通纹理分解和无下采样轮廓波变换(NSCT)的卫星云图边缘特征提取算法。首先使用基于张量扩散的卡通纹理分解对卫星云图进行预处理,减小云图中噪声和纹理的影响,将云的缓变边缘转化为阶跃边缘;然后依据NSCT域中高频系数的正负关系和相关性进行边缘特征提取。通过实验,将文中算法分别与基于投影卡通纹理分解的预处理算法、小波多尺度积、NSCT模极大值和NSCT多尺度积等边缘特征提取算法进行比较。结果表明,本算法可更准确地提取云的边缘,且受噪声和纹理的影响小,边缘的连续性更好,为准确地进行云分类提供了保障。

Edge extraction of satellite cloud images based on cartoon texture decomposition and non-subsampled contourlet transform

For the convenience of the cloud edge extraction in a cloud image,an algorithm for edge feature extraction of satellite cloud images based on cartoon texture decomposition and non-subsampled contourlet transform (NSCT) was proposed in this paper. Firstly, the satellite cloud images were preprocessed by using cartoon texture decomposition based on diffusion tensor. The effects of noise and textures were reduced, and the gradually changing edges were turned into step edges. Then edge features were extracted according to the relationship of the polarity and correlation of high-frequency coefficients in NSCT domain. Through a large number of experiments, the proposed algorithm was correspondingly compared with the preprocessing algorithm of cartoon texture decomposition based on projection and the edge feature extraction algorithms such as wavelet multi-scale product, NSCT modulus maxima, NSCT multi-scale product, respectively. The experimental results show that the proposed algorithm can extract the cloud edges more accurately. The extracted edges have not only better continuity, but also fewer effects of noise and textures. These provide guarantee for the subsequent accurate classification of clouds in satellite images.