“Low-rank + dual” model based dimensionality reduction

详细信息    查看全文

• 作者：Si-Qi Wang ; ^{pearlwangxd@163.com" class="auth_mail" title="E-mail the corresponding author}Author Vitae ; Xiang-Chu Feng ^{xcfeng@mail.xidian.edu.cn" class="auth_mail" title="E-mail the corresponding author}Author Vitae ; Wei-Wei Wang ^{wwwang@mail.xidian.edu.cn" class="auth_mail" title="E-mail the corresponding author}Author Vitae
• 关键词：Dimensionality reduction ; Background modeling ; Singular value decomposition ; Thresholding method ; lp-Minimization problem
• 刊名：Neurocomputing
• 出版年：2016
• 出版时间：20 February 2016
• 年：2016
• 卷：178
• 期：Complete
• 页码：3-10
• 全文大小：906 K

文摘

We propose a novel “low-rank + dual” model for the matrix decomposition problems. Based on the unitarily invariant property of the Schatten p  -norm, we prove that the solution of the proposed model can be obtained by an “16173&_mathId=si0002.gif&_user=111111111&_pii=S0925231215016173&_rdoc=1&_issn=09252312&md5=e7aff2bd5bc2e0e96d86733d93624339" title="Click to view the MathML source">l_∞+l₁$l_{\infty }+l_1$” minimization problem, thus a simple and fast algorithm can be provided to solve our new model. Furthermore, we find that applying “16173&_mathId=si0002.gif&_user=111111111&_pii=S0925231215016173&_rdoc=1&_issn=09252312&md5=e7aff2bd5bc2e0e96d86733d93624339" title="Click to view the MathML source">l_∞+l₁$l_{\infty }+l_1$” to any vector can achieve a shifty threshold on the values. Experiments on the simulation data, the real surveillance video database and the Yale B database prove the proposed method to outperform the state-of-the-art techniques.