基于超像素的互惠最近邻聚类彩色图像分割

Color image segmentation based on superpixels and reciprocal nearest neighbors clustering

为了提高图像分割质量,提出一种基于超像素的互惠最近邻聚类彩色图像分割方法。该方法先将图像进行过分割处理构成超像素集,超像素代替原图像像素点作为互惠最近邻聚类的输入数据,以超像素的均值距离和方差融合相邻超像素的边界信息构成一种新的相似度度量,利用互惠最近邻聚类算法实现图像的区域融合。实验证明,该方法不仅实现图像自动分割,并且获得较好的分割质量。

To improve the quality of image segmentation, a color image segmentation method based on superpixels and reciprocal nearest neighbor （RNN） clustering is presented. Firstly, a color im- age is converted into superpixels through over-segmentation. Then image region merging is performed by RNN clustering algorithm to superpixels which are stead of original image pixels, taking a new similarity regarded as the input of RNN clustering, in- measurement with average distance and vari- ance combined boundary information of neighboring superpixels. Experimental results show that the method not only realizes image automatic segmentation but also obtains a good segmentation quality.