遥感图像的异质性测度分割效果评价

Heterogeneity Measure Based Segmentation Performance Evaluation for Remote Sensing Image

针对高分辨率遥感图像分割结果的评价问题,提出一种基于异质性测度的非监督分割评价方法。首先,通过全局方差和加权Moran指数分别表示对象内异质性和对象间异质性,并利用二者归一化后的和式来对整体分割结果进行评价。其次,为了进行局部分割结果评价,提出一种基于对象方差和局部Geary指数的异质性测度。最后,利用多分辨率分割方法对遥感图像进行分割,并利用提出的方法进行分割评价。实验结果表明,提出的方法能够对不同分割尺度结果进行有效评价,同时可以对过分割区域和欠分割区域进行判断。

In order to evaluate segmentation quality of high resolution remote sensing image, an un-supervised segmentation evaluation method based on heterogeneity measure was proposed. Firstly, global variance and weighted Moran index were introduced to express the intro-object and inter-object heterogeneity. Then the two heterogeneity measure were normalized and summed to evaluate the whole performance of segmentation result. Secondly, to evaluate the local quality of image objects, a heterogeneity measure based on object variance and local Geary index was presented. Finally, an experiment is carried out on a remote sensing which was segmented by multi-resolution segmentation method. And heterogeneity measure proposed in this paper was used to evaluate the segmentation result.It shows that the heterogeneity measure can effectively evaluate the different scale segmentation results and meanwhile can identify regions which are over-segmented or under-segmented.