Remote Sensing Image Segmentation Based On K-means

Remote sensing image segmentation has played a vital and significant role inpeople’s growing demand for remote sensing data processing, while the complex、multi-level、 multi-spectrum or other features makes remote sensing imagesegmentation one of the most important and difficult topics in remote sensing imageprocessing field. Although scholars has did much extensive and in-depth studies inimage segmentation，image segmentation applications is still facing great challengedue to various reasons, especially for the development of remote sensing imagetechnology.Based on the principles of the K-means Clustering algorithm、Otsu method andlevel-set method, two remote sensing image segmentation model are proposed, andthe experimental results are summarized analyzed. The main work and contributionsof this paper are as follows.First, the theory of image segmentation is introduced, the author classify theexisting image segmentation methods into four categories and give them a systematicexposition separately, mainly present the purpose and significance of this research.Second, a new remote sensing image segmentation model combines the Otsumethod and K-means clustering algorithm is proposed considering their merits anddefects. Several study cases are conducted to show the application of this method tothe remote-sensing image segmentation, and a large number of experiment resultsshow that the proposed method not only overcomes the incomplete information of thetraditional Otsu algorithm and enhances its robustness, but also obtains anoptimization of the K-means clustering algorithm. The proposed algorithm achievesimprovements on both of computing speed and segmentation performance.Third, another remote sensing image segmentation model has been implementedbased on K-means algorithm also with the improved multi-phrase level set model.Comparing with the classical multi-phase C-V model, the improved model considers the region area information, gradient information and edge detection. Experimentalresults show that the proposed approach has advantages in rapid and efficientapplication for remote sensing image segmentation.