多尺度归一化割用于遥感图像分割

摘要

Due to high-resolution remote sensing images have characteristics of large size,rich feature types,and complex texture,we use multi-scale normalized cut method for high resolution remote sensing image segmentation.The method is a global segmentation method based on graph theory.The method is able to take into account information of coarse and fine layers,thus it is suitable for complex feature segmentation.We use Ikonos image of a region of Nanjing to conduct experiments, and to compare it with multi-scale segmentation module in mainstream segmentation software eCognition.The evaluation results show that the method can achieve sound results for high-resolution remote sensing image segmentation,which has good research prospects.%针对高分辨率遥感图像数据量大、地物类型丰富、纹理结构复杂等特点，该文利用多尺度归一化割方法进行高分辨率遥感图像的分割。该方法是一种从全局出发的基于图论的分割方法，能够兼顾粗糙层和精细层的信息，适合复杂地物的分割。利用南京某区域 Ikonos 数据进行了实验，并与主流遥感图像分割软件 eCognition 中多尺度分割模块进行了效果对比和定量评价。结果表明，该方法在高分辨率遥感图像分割中可以取得较好的效果，具有较好的研究前景。