基于灰度和空间特性的谱聚类图像分割

Spatial and gray feature-based spectral clustering for image segmentation

为了克服谱聚类图象分割方法性能容易受到图像大小和相似性测度的影响,提出一种基于灰度和空间特性的谱聚类图像分割算法。该算法不对图像中的像素之间建立相似性,而是利用各个像素的灰度在图像中的分布信息和像素点的空间邻接信息建立灰度之间的相似关系,通过对图像中灰度的分类进而获得原始图像的分割结果。因此,该算法不会受到图像大小的限制,无论对于多大的图像,相似性矩阵的大小都是小于等于256×256。Berke-ley基准图像数据集上的分割仿真实验验证了该方法的有效性。

To overcome the influence of the image size and similarity measure to the performance of spectral clustering,a novel spatial and gray feature-based spectral clustering algorithm for image segmentation is proposed.It introduces a function called spatial-gray compactness to construct the similarity relationship between any two grays,not between any two pixels.The method utilizes the distribution of the gray and the spatial adjacency of the pixel in the image to classify the gray levels,and eventually performs the classification of the pixels.No matter what the image size is,the size of the obtained similarity matrix is smaller than 256×256.Experimental results on the Berkeley segmentation dataset and benchmark show that the novel method is effective.