Steven C.H. Hoi and Michael R. Lyu

1
The Chinese University of Hong Kong
A Semi-Supervised Active Learning Framework for
Image Retrieval
Steven C.H. Hoi and Michael R. Lyu Department of
Computer Science and Engineering The Chinese
University of Hong Kong, Shatin, N.T., Hong Kong
SAR chhoi, lyu_at_cse.cuhk.edu.hk
Motivations
Contributions
1
2

• The semantic gap between low-level features and
  high-level concepts is challenging for image
  retrieval.
• Unlabeled data may be beneficial for bridging
  the semantic gap.
• The computation cost is challenging for
  unlabeled data.
• Little work in image retrieval is effective and
  
• efficient for learning with unlabeled data.

• A novel framework is proposed for learning with
  labeled and unlabeled data which comprises a
  fusion of semi-supervised learning and SVM.
• Based on the proposed framework, a novel active
  learning algorithm is proposed for image
  retrieval.
• Promising experimental results show our
  proposed framework and algorithm are effective.

.
Architecture
3
4
Algorithm
5
Formulation and Analysis
SVM
SSL
Experimental Results
6
Experimental results on the 20-Cat dataset
Conclusion
The proposed semi-supervised active learning
framework and algorithms are effective and very
promising for learning with labeled and
unlabeled data in image retrieval.
Dept. of C.S. E., The Chinese University of
Hong Kong
IEEE Computer Vision and Pattern Recognition 2005