针对当前地震后震区时空数据难以及时获取的问题,采用了一种改进均值漂移(Mean Shift)信息提取方法。首先,将原始影像划分为纹理区和均色区。均色区域直接利用Mean Shift算法获得; 纹理区域则利用归一化分布密度值获取合适的带宽,再使用Mean Shift算法进行信息提取。通过构造代价函数判别相邻区域是否需要合并,以消除过分割区域。最后,提出了一种信息提取匹配指数对信息提取结果进行评价,并将该方法应用于汶川和芦山地震后获取的无人机高空间分辨率影像,进行信息提取实验。实验结果表明:所提出的改进Mean Shift算法提取精度优于传统Mean Shift算法,为后续的地震灾情评估提供了基础数据。
In order to solve the problem that the spatial-temporal data is difficult to be obtained in earthquake-stricken area after the earthquake, we proposed an improved Mean Shift algorithm for information extraction. Firstly, the image was divided into texture areas and homogeneous color areas through different characteristic of variance detection on the color space. Preliminary partition on the homogeneous color area was directly achieved by Mean Shift algorithm. Meanwhile, for the texture area, a high-dimensional feature space was set up by extracting the shape, texture and color information, and the proper bandwidth was calculated according to the normalized distribution density, then the Mean Shift algorithm was applied on the feature space for model classification to reach the partition. Secondly, a cost function was set up to realize weather the adjacent area needed to be merged to smooth over the partitioned area. Finally, an information extraction matching index(EMI)which considering the area and spectrum was proposed to evaluate the extraction results. The test results for high spatial resolution remote sensing image information extraction on UAV images were conducted in Lushan and Wenchuan earthquake-stricken areas. The experimental results show that the improved Mean Shift algorithm performs better than the original ones, and provides data protection for damage information extraction of subsequent earthquake.
