基于L21范式的多图正则化非负矩阵分解方法

针对非负矩阵分解方法对原始数据的单图约束导致的结果未知性大、满足需求单一,以及大多非负矩阵分解方法存在对噪声、离群点较敏感导致的稀疏度和鲁棒性较差等问题,提出基于L21范式的多图正则化非负矩阵分解方法。采用L21范式,提升分解结果的稀疏度和鲁棒性。构建多图约束的算法模型更好地保持数据的流形结构。构建目标函数并给出乘性迭代规则。通过在多个数据库上的实验表明,该方法在识别效果上有明显的提升。

MULTIPLE GRAPH REGULARIZED NON-NEGATIVE MATRIX FACTORIZATION BASED ON L21 NORM

The non-negative matrix factorization method,the single graph constraint of the original data results in large unknowns,satisfying single demand,and most non-negative matrix factorization methods have poor sparsity and robustness against noise and outliers.A multi-graph regularized non-negative matrix factorization method based on the L21 norm is proposed.The L21 norm was used to improve the sparsity and robustness of the decomposition results.We constructed multi-graph constraints to better maintain the manifold structure of the data,and we constructed the objective function and gave a multiplication iteration rule.The experiments on multiple databases show that the proposed method has a significant improvement in recognition performance.