多维特征自适应MeanShift遥感图像分割方法

An Adaptive MeanShift Segmentation Method of Remote Sensing Images Based on Multi-Dimension Features

针对MeanShift算法分割遥感图像的自动化程度和精度不高的问题,提出一种多特征自适应Mean-Shift遥感图像分割方法。3组实验结果表明,本方法相比EDISON软件能得到更好的分割效果,且能在一定程度上提高遥感影像分割的自动化。

Due to low segmentation efficiency and low accuracy of Mean-Shift algorithm, thispaper puts forward to an adaptive Mean-Shift segmentation method of remote sensing ima- ges. Firstly, location features, multi band spectrum principal components and texture fea- tures are extracted to form multi-dimension feature spaces. Then, based on classical Mean Shift clustering algorithm, initial clustering images are got by using less fixed space band- widths and global optimal spectrum bandwidths that are estimated by plug-in rules. Mean space distance, mean spectrum distance and texture distance are calculated for each region in the initial clustering images, and used for space bandwidths, spectrum bandwidths and tex- ture bandwidths of sequential clustering. Further, multi-dimension feature Mean-Shift Clus- tering was implemented by using calculated bandwidths. Lastly, the clustered regions are combined to get segmentation images. Three experiment results of remote sensing images show that the proposed method in this paper are better than EDISON software, and to some extent improve the efficiency of segmentation of remote sensing images.