一种改进的谱聚类彩色图像分割方法

A Color Image Segmentation Approach Based on Improved Spectral Clustering

提出一种改进的基于谱聚类的彩色图像分割方法,首先引入Levin's Affinity的权函数代替传统的高斯核函数建立相似矩阵来构造带权无向图,从而更精细地刻画出数据间的特征相似性;其次,采用线性映射将图嵌入到一个由部分特征向量生成的子空间中,使得数据映射到新的空间后也能较好的保留其在原空间中的结构;最后,在生成的子空间中用K均值聚类算法进行聚类从而为每个像素点分配类标签达到彩色图像分割的目的.与相关谱聚类算法进行图像分割的结果比较证实了改进算法的有效性和显著性.

A color image segmentation approach based on improved spectral clustering is proposed. First, the weighted function of a weighted undirected graph is defined by Levin＇s Affinity which can better capture the feature similarity between data than the tradition- al Ganssian kernel function. Secondly, the graph can be embedded into a subspace that contains a part of eigenvectors through liner mapping, which keeps the structure in the original space successfully when the data is mapped into a new space. Finally, K-means clustering algorithm is used in the subspace to distribute labels for each pixel and achieve the segmentation of color images. The ex- periment results that compared with the related spectral clustering algorithms are provided to show the effectiveness and significance of the improved algorithm.