基于形态学多重分形的遥感图像多尺度分割  

Multiscale segmentation of remote sensing images based on local morphological multifractal exponents

作　　者：夏勇[1] 赵荣椿[1] 

机构地区：[1]西北工业大学计算机学院,陕西西安710072

出　　处：《计算机应用》2006年第9期2071-2073,共3页journal of Computer Applications

基　　金：国家自然科学基金资助项目(60141002);航空基金资助项目(02I53073)

摘　　要：提出了一种基于数学形态学的局部多重分形指数特征来描述图像中的纹理信息,并构造了基于图像四叉树的多尺度分割算法来实现遥感图像的粗分割。形态学多重分形指数能够准确而全面的刻画纹理的局部尺度特性,而多尺度分割算法可以在保持分割精度的前提下大大降低时间复杂度。在遥感图像上进行的对比实验表明,该算法在分割的效果和效率上都优于使用其他纹理特征的分割算法。To separate artificial regions in a remote sensing image from natural background, a new texture descriptor named the LMME（Local Morphological Multifractal Exponents） and a quadtree-based multiscale segmentation algorithm were proposed in this paper. To assess its performance in segmentation of remote sensing images, the proposed approach was compared with four other approaches. The formerused box-counting based multifractal dimensions, and the latter used a Ganssian Markov random field based feature. The experimental results demonstrate that the proposed approach can provide more effective and more efficient segmentations.

关 键 词：图像分割 多重分形估计 数学形态学 多尺度分割 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]