CAJ | 학술논문

基于图正则化非负矩阵分解算法(GNMF)，提出一种基于凸光滑的L3/2范数正则化图非负矩阵分解算法。该算法用非负矩阵分解算法对数据进行低维非负分解时，根据流形学习的图框架理论，构建邻接矩阵保持数据局部几何结构，并对数据的低维表示特征进行凸光滑的L3/2范数稀疏性约束，在给出算法更新迭代规则的同时，从理论上证明了所给算法的收敛性。通过人脸数据库 ORL、手写体数据库USPS和图像库COIL20的仿真实验表明，相对于非负矩阵分解算法及其基于稀疏表示的改进算法，所给算法均具有更高的聚类精度。
기우도정칙화비부구진분해산법(GNMF)，제출일충기우철광활적L3/2범수정칙화도비부구진분해산법。해산법용비부구진분해산법대수거진행저유비부분해시，근거류형학습적도광가이론，구건린접구진보지수거국부궤하결구，병대수거적저유표시특정진행철광활적L3/2범수희소성약속，재급출산법경신질대규칙적동시，종이론상증명료소급산법적수렴성。통과인검수거고 ORL、수사체수거고USPS화도상고COIL20적방진실험표명，상대우비부구진분해산법급기기우희소표시적개진산법，소급산법균구유경고적취류정도。
This paper presents a novel algorithm called L3/2 regularized graph non-negative matrix factorization,which was based on the convex and smooth L3/2 norm.When original data is factorized in lower dimensional space by non-negative matrix factorization,L3/2 regularized graph non-negative matrix factorization preserves the local structure and intrinsic geometry of data,with the aid of the convex and smooth L3/2 norm as sparse constrain for the low dimensional feature.An efficient multiplicative updating procedure was produced along with its theoretic j ustification of the algorithm convergence.Compared with non-negative matrix factorization and its improved algorithms based on sparse representation,the proposed method achieves better clustering results,which is shown by experiment results on ORL face database,USPS handwrite database,and COIL20 image database.