视频前背景分离论文之（3） Group Sparsity in Nonnegative Matrix Factorization

1、Abstract

（1）features or data items belonging to the same group share the same sparsity pattern in low-rank factors.
（2）Proposed mixed-norm regularization to promote group sparsity in the factor matrices of NMF.

2、NMF Model

$$f(W,H)=\frac{1}{2} \Vert X-WH \Vert_F^2$$
where $W\in \mathbb R_+^{m\times k}$ and $H\in \mathbb R_+^{k\times n}$.

Group Sparsity in NMF：

3、GSNMF

GSNMF Model:

$$f(W,H)=\frac{1}{2}\sum_{b=1}^{B}\Vert X^{(b)}-WH^{(b)}\Vert_F^2$$
where the columns of $X\in\mathbb R^{m\times n}$ are dived into $B$ groups.
$$\min_{W\ge0,H\ge0}f(W,H)+\alpha\Vert W\Vert_F^2+\beta \sum_{b=1}^B\Vert H^{(b)}\Vert_{1,q}$$

The ${l_{1,q}-\text{norm}}$ of $Y\in\mathbb R^{a\times c}$ is defined by

$$\Vert Y\Vert_{1,q}=\sum_{j=1}^a\Vert \mathbf y_{j\cdot}\Vert_q$$

Group Sparsity in NMF：