摘要
颜色迁移是组织病理学图像颜色预处理中的重要环节.为了解决颜色迁移过程中某些重要结构颜色改变的问题,在保结构颜色迁移(structure-preserving color normalization,SPCN)算法基础上融合聚类过程,并结合稀疏非负矩阵分解(sparse non-negative matrix factorization,SNMF)提出K均值稀疏非负矩阵分解基组合(K-means and SNMF basis combination,KSBC)算法.首先通过K均值算法对图像聚类,根据聚类中心识别细胞结构;然后求解稀疏非负矩阵分解模型得到染色基和结构矩阵,根据聚类结果对结构矩阵和染色基准确组合.KSBC算法承袭了SPCN算法的特性,又能灵活地迁移和保留原图像结构颜色.在组织病理学图像数据库中进行对比实验,KSBC算法在图像质量评估指标上优于直方图匹配,Reinhard,Macenko,SPCN和高阶矩算法,并提高残差神经网络的泛化性能.
Color transfer is an important process in histopathological color normalization.In order to preserve structural color,this paper proposes KSBC(K-means and SNMF basis combination)algorithm based on SPCN(structure-preserving color normalization)algorithm with clustering progress and sparse non-negative matrix factorization.Firstly,KSBC algorithm clusters image by K-means algorithm,and identifies cell structure based on cluster centers.Then we solve sparse non-negative matrix factorization model to obtain stain vectors and structure matrix,accurately combining structure matrix with the corresponding stain vectors according to the clustering results.KSBC algorithm inherits the characteristics of SPCN algorithm,and can flexibly transfer and preserve the structural color of the original image.This paper performs comparative experiments in histopathology image database,and shows that the quality evaluation metrics of image of KSBC algorithm is better than histogram specification,Reinhard,Macenko,SPCN,and higher moments algorithms,and improves residual neural network generalization performance.
作者
张术昌
袁梓洋
王红霞
陈波
Zhang Shuchang;Yuan Ziyang;Wang Hongxia;Chen Bo(Department of Mathematics,College of Arts and Sciences,National University of Defense Technology,Changsha 410073)
出处
《计算机辅助设计与图形学学报》
EI
CSCD
北大核心
2020年第12期1890-1897,共8页
Journal of Computer-Aided Design & Computer Graphics
基金
国家自然科学基金(61977065).
