摘要
目的提出一种改进的自适应谱聚类图像分割算法,该算法能自动选择出最优尺度参数从而提高谱聚类算法分割的准确率。方法利用约束条件优化相关准则函数,对相似度量函数自动学习迭代并得到最优尺度参数,再运用基于Nystr m估计的谱聚类算法得到最后的图像分割结果。选择对不同性质的纹理图像采用适合的相似度量函数并应用本文的算法进行图像分割,最后与k-均值算法和预分割后再使用人工调整到最优参数的谱聚类算法的分割结果进行了比较。结果这种改进的自动选择最优尺度参数的谱聚类算法在分割效果上较其它两种聚类算法能得到更好的分割结果。结论本文提出的改进方法,能使谱聚类算法的图像分割效果更理想。
Objective To propose an improved adaptive spectral clustering method for image segmentation to allow automatic selection of the optimal scaling parameters and enhance the accuracy of spectral clustering.Methods Using constrain conditions for optimizing the criterion function and determining the optimal scaling parameters by iteration,the final image segmentation was achieved through spectral clustering based on Nystr？m approximation.We chose suit weight functions for different texture images,and used the proposed method for image segmentation.The k-means algorithm and the method of spectral clustering after pre-segmentation by manually choosing the scaling parameter were compared with the proposed method.Results The improved spectral clustering algorithm with automatic selection of the optimal scaling parameters achieved better results of image segmentation than the other two methods.Conclusion The proposed algorithm can improve the accuracy of spectral clustering for image segmentation.
出处
《南方医科大学学报》
CAS
CSCD
北大核心
2012年第5期655-658,663,共5页
Journal of Southern Medical University
基金
国家自然科学基金(81000642)
关键词
谱聚类
图像分割
自适应
Nystrm估计
spectral clustering
image segmentation
adaptive
Nystrom approximation
