基于加权非负矩阵分解的非负张量分解算法  

Non-negative Tensor Factorization Based on Weighted Non-negative Matrix Factorization

作　　者：刘路路[1] 刘亚楠[1] 王凡[1] 

机构地区：[1]合肥师范学院计算机科学与技术系,安徽合肥230601

出　　处：《合肥师范学院学报》2016年第3期24-27,共4页Journal of Hefei Normal University

基　　金：安徽省高校省级优秀青年人才基金重点项目(2011SQRL129ZD);合肥师范学院科研项目(2015QN15)

摘　　要：提出一种基于加权非负矩阵分解的非负张量分解算法。为了充分利用图像本身的结构信息与内在几何结构,首先根据图像类别构造权值矩阵,把图像集合构造成三阶张量,然后,针对该三阶张量利用张量几何运算与非负矩阵分解得到非负张量分解算法的初值,最后实现图像的分类。实验结果表明该算法应用于手写数字图像库中能有效改善图像分类的准确性。This paper proposes a non-negative tensor factorization based on weighted non-negative matrix factorization.In order to fully use the structural information of the data and their intrinsic geometry,the weighted matrix is constructed according to the label of images and the set of images is constructed to a third order tensor first.Then using tensor geometry operation and non-negative matrix factorization can get the original values.Finally,image classification is realized.Experimental results on the handwritten digital image database show that the proposed algorithm can effectively improve the accuracy of the image classification.

关 键 词：图像分类 非负矩阵分解 非负张量分解 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]