Abstract: In this paper, a method of image segmentation is presented, which based on fast global K-means and region merging. Medial filter is used to remove the noise of target image. The initial segmented result is obtained by using fast global K-means clustering algorithm in the color space. A region merging strategy is used to merge the initial regions with the goal of forming the final segmentation result. The simulation results indicate that compared with other methods, the segmentation result is well consistent with human perception, especially in image details.

Key words: image segmentation, fast global K-means algorithm, region merging, clustering analysis

摘要： 提出一种融合快速全局K-means与区域合并的图像分割方法。该方法利用中值滤波方法对图像去噪；运用快速全局K-means算法对图像的颜色空间进行聚类分析；结合区域合并准则，对初始分割合并得到最终的分割结果。实验表明，与同类算法比较，该方法的分割结果在图像细节方面能够很好地满足人的主观视觉。

关键词: 图像分割, 快速全局K-means, 区域合并, 聚类分析