摘要
为了有效挖掘序列数据的时空信息,提出联合l p和l 2,p范数极小化的序列子空间聚类算法.首先,定义依赖于样本距离的权重,构造基于l 2,p范数的时序图,刻画数据在时间维度上的局部相似性.然后,考虑到非凸l p 0<p<1范数最小化通常结果优于凸的l 1范数,能更有效地切断语义无关数据间的联系,所以采用l p范数度量表示矩阵的稀疏性.最后,通过线性化交替方向法求解优化模型.在视频、运动、人脸数据上的实验表明文中算法的有效性.
To extract the spatio-temporal information in sequential data effectively,a sequential subspace clustering method via joint l p/l 2,p-norms minimization is proposed.Firstly,a l 2,p-norm temporal graph is constructed to describe local similarity along the temporal direction by defining the sample-distance dependent weights.Secondly,since non-convex l p-norm(0<p<1)minimization delivers better results than convex l 1-norm minimization does,and it removes more links between semantically-unrelated samples,l p-norm is adopted to measure the sparsity of representation matrix.Finally,the linearized alternating direction method is employed to solve the optimization model.Experiments on video dataset,motion dataset,and face dataset confirm the effectiveness of the proposed method.
作者
胡文玉
李声豪
涂志辉
易云
HU Wenyu;LI Shenghao;TU Zhihui;YI Yun(School of Mathematics and Computer Science,Gannan Normal University,Ganzhou 341000)
出处
《模式识别与人工智能》
EI
CSCD
北大核心
2020年第3期221-233,共13页
Pattern Recognition and Artificial Intelligence
基金
国家自然科学基金项目(No.61863001,61962003,11661007,61702244,11761010)
江西省自然科学基金项目(No.20181BAB202021,20192BAB205086)
江西省研究生创新专项资金项目(No.YC2019-S393)
赣南师范大学科研项目(No.18zb04,YCX18B001)资助。
关键词
子空间聚类
序列数据
稀疏编码
低秩表示
Subspace Clustering
Sequential Data
Sparse Coding
Low-Rank Representation