Proposed Scheme

1
The Chinese University of Hong Kong
Web Image Learning for Searching Semantic
Concepts in Image Databases
Chu-Hong 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
Introduction
Architecture
Without textual descriptions or label information
of images, searching semantic concepts in image
databases is still a very challenging task. This
poster presents a scheme to learn web images for
searching semantic concepts from image databases.
To formulate effective algorithms, Support Vector
Machines are engaged to help the learning tasks.
Experimental results show that the proposed
scheme is effective and promising.

• The system consists of the following modules
  shown in the following figure
• The first step is to search the Web images from
  WWW by the input keywords.
• From the returned Web images, clustering learning
  is applied and the training sets are obtained
  after removing noisy images.
• Based on the training sets, relevance feedback is
  suggested for interactive learning using Support
  Vector Machine techniques.

Proposed Scheme

• Searching and Clustering
• Users typing the keywords to describe the desired
  semantic concepts
• Searching related Web images associated with the
  keywords from WWW
• Clustering the searching results by the k-means
  algorithm
• Removing the noisy images to obtain the final
  training sets of web images
• Relevance Feedback Learning by SVMs
• SVM provides good generalization performance and
  very excellent results on pattern classification
  problems.
• Preliminary Learning employing one-class SVMs
  since only positive training
• samples are available.
• Relevance Feedback Learning engaging two-class
  SVMs for learning iteratively.

Demo- Searching fireworks
1
Searching by keywords from WWW
2
Clustering results
Cluster1
Cluster2
3
4
Results before relevance feedback
Results after 3-round feedbacks
Experimental Results
Dept. of Computer Science and Engineering,
C.U.H.K.
International World Wide Web Conference 2004