改进的谱聚类算法在图像分割中的应用

为了消除传统的谱聚类图像分割算法存在的缺陷,提出一种改进的谱聚类图像分割算法.该算法提出余弦相似性加权矩阵,充分利用图像的纹理信息和空间临近信息构造相似性矩阵.在谱映射过程中,利用Nystr(o)m逼近策略估计相似性矩阵及其主特征向量.最后利用优化的K-means算法与优化的粒子群算法相结合的算法对得到的低维向量子空间进行聚类,避免直接采用K-means算法对初始值敏感,易陷入局部最优的缺点.实验证明该算法在运行时间和分割精度方面较传统谱聚类算法均有明显的提高.

Application of Improved Spectral Clustering Algorithm in Image Segmentation

In order to eliminate the defects of traditional spectral clustering image segmentation algorithm,an improved spectral clustering image segmentation algorithm was proposed,which made full use of the image texture information and spatial adjacency information to construct cosine similarity matrix.In the spectral mapping process,the similarity matrix and its main eigenvectors were estimated by using the Nystr(o)m approximation strategy.Finally,a new algorithm combining improved K-means and optimized particle swarm optimization algorithm was used to cluster the low-dimensional subspace,which avoided the K-means algorithm being sensitive to the initial value and easy to fall into the local optimum.Experimental results show that the new method has obviously better performance and low computational cost than the traditional spectral clustering algorithm.