An Adaptive Semisupervised Feature Analysis for Video Semantic Recognition

被引:283
作者
Luo, Minnan [1 ]
Chang, Xiaojun [2 ]
Nie, Liqiang [3 ]
Yang, Yi [4 ]
Hauptmann, Alexander G. [2 ]
Zheng, Qinghua [1 ]
机构
[1] Xi An Jiao Tong Univ, Dept Comp Sci, SPKLSTN Lab, Xian 710049, Shaanxi, Peoples R China
[2] Carnegie Mellon Univ, Sch Comp Sci, Pittsburgh, PA 15213 USA
[3] Shandong Univ, Sch Comp Sci & Technol, Jinan 250100, Shandong, Peoples R China
[4] Univ Technol Sydney, Ctr Quantum Computat & Intelligent Syst, Sydney, NSW 2007, Australia
基金
美国国家科学基金会;
关键词
Feature selection; manifold regularization; semisupervised learning; video semantic recognition; FEATURE-SELECTION;
D O I
10.1109/TCYB.2017.2647904
中图分类号
TP [自动化技术、计算机技术];
学科分类号
0812 ;
摘要
Video semantic recognition usually suffers from the curse of dimensionality and the absence of enough high-quality labeled instances, thus semisupervised feature selection gains increasing attentions for its efficiency and comprehensibility. Most of the previous methods assume that videos with close distance (neighbors) have similar labels and characterize the intrinsic local structure through a predetermined graph of both labeled and unlabeled data. However, besides the parameter tuning problem underlying the construction of the graph, the affinity measurement in the original feature space usually suffers from the curse of dimensionality. Additionally, the predetermined graph separates itself from the procedure of feature selection, which might lead to downgraded performance for video semantic recognition. In this paper, we exploit a novel semisupervised feature selection method from a new perspective. The primary assumption underlying our model is that the instances with similar labels should have a larger probability of being neighbors. Instead of using a predetermined similarity graph, we incorporate the exploration of the local structure into the procedure of joint feature selection so as to learn the optimal graph simultaneously. Moreover, an adaptive loss function is exploited to measure the label fitness, which significantly enhances model's robustness to videos with a small or substantial loss. We propose an efficient alternating optimization algorithm to solve the proposed challenging problem, together with analyses on its convergence and computational complexity in theory. Finally, extensive experimental results on benchmark datasets illustrate the effectiveness and superiority of the proposed approach on video semantic recognition related tasks.
引用
收藏
页码:648 / 660
页数:13
相关论文
共 53 条
  • [1] Aggarwal CC, 2001, LECT NOTES COMPUT SC, V1973, P420
  • [2] Soft-constrained Laplacian score for semi-supervised multi-label feature selection
    Alalga, Abdelouahid
    Benabdeslem, Khalid
    Taleb, Nora
    [J]. KNOWLEDGE AND INFORMATION SYSTEMS, 2016, 47 (01) : 75 - 98
  • [3] [Anonymous], 2011, P 1 INT C MULT RETR
  • [4] [Anonymous], 2006, BOOK REV IEEE T NEUR
  • [5] [Anonymous], 2012, ELEMENTS INFORM THEO
  • [6] [Anonymous], 2000, Pattern Classification, DOI DOI 10.1007/978-3-319-57027-3_4
  • [7] [Anonymous], 2008, Advances in Neural Information Processing Systems
  • [8] [Anonymous], 2009, Tech. Rep. CMU-CS- 09-161
  • [9] [Anonymous], Learning from Labeled and Unlabeled Data with Label Propagation
  • [10] [Anonymous], 1998, Feature Extraction, Construction and Selection: A Data Mining Perspective