Locally non-negative linear structure learning for interactive image retrieval

ABSTRACT

A successful interactive image retrieval system is expected to quickly return as many relevant results as possible while costing less users' effort. Considering these system demands, firstly we propose a novel semi-supervised learning algorithm called Locally Non-negative Linear Structure Learning (LNLS), which is based on the assumption that the labels of each data should be sufficiently smooth with respect to the locally non-negative linear structure of dataset. It has two main merits: first, it is robust to the small sample learning problem since it learns structure from both labeled and unlabeled data; second, by emphasizing the non-negativity of locally linear structure, this algorithm preserves the non-negative inherent characteristic of image data and can truly reveal the intrinsic structure of the images corpus, especially the asymmetric relationship between images. Meanwhile, we explore an online updating algorithm for LNLS to tackle the large computation cost. Thus the model can be generalized to the new queries or the newly-labeled samples without retraining. Furthermore, an active learning method for LNLS is proposed to make the most of users' effort to improve the learner. The encouraging experimental results demonstrate the effectiveness and efficiency of our proposed methods.

REFERENCES

• 1: Thomas S. Huang, et al. Active learning for interactive multimedia retrieval. In Proc. of the IEEE, vol. 96, no. 4, 648--667, 2008.
• 2: X. Zhu, et al. Semi-supervised learning using gaussian fields and harmonic functions. In Proc.of Int'l Conf. on Machine Learning, 2003.
• 3: D. Zhou, O. Bousquet, T. N. Lal, J. Weston, and B. Scholkopf. Learning with local and global consistency. In Proc. of Advances of Neural Information Processing, 2004.
• 4: J.R. He, et al. Manifold--Ranking Based Image Retrieval. In. Proc. of ACM Multimedia, 2004.
• 5: M. Wang, et al. Video annotation by graph-based learning with neighborhood similarity. In. Proc. of ACM Multimedia, 2008.
• 6: S.t. Roweis and L.K. Saul. Nonlinear dimensionality reduction by locally linear embedding. Science: vol. 290, no. 5500, 2323--2326. 2000.
• 7: F. Wang, et al. Semi-Supervised classification using linear neighborhood propagation. In. Proc. of IEEE Conf. on Computer Vision and Pattern Recognition, 2006.
• 8: D.D. Lee, H.S. Seung. Learning the parts of objects by non-negative matrix factorization. Nature: vol. 401, 788--791. 1999
• 9: S. Tong and E. Chang. Support vector machine active learning for image retrieval. In. Proc. of ACM Multimedia, 2001
• 10: J. Tang, et al. Structure-Sensitive Manifold Ranking for Video Concept Detection. In. Proc. of ACM Multimedia, 2007

INDEX TERMS

Primary Classification:
  H. Information Systems
  H.3 INFORMATION STORAGE AND RETRIEVAL
      H.3.3 Information Search and Retrieval
          Subjects: Search process

Additional Classification:
  H. Information Systems
  H.3 INFORMATION STORAGE AND RETRIEVAL
      H.3.3 Information Search and Retrieval
          Subjects: Relevance feedback

General Terms:
Algorithms, Experimentation

Keywords:
active learning, interactive image retrieval, locally non-negative linear structure