Relevance Feedback In ContentBased Image Retrieval System KienPing Chung Supervisor: A' Prof' Fung S

1
Relevance Feedback In Content-Based Image
Retrieval SystemKien-Ping Chung Supervisor A.
Prof. FungSchool of Information Technology,
Murdoch University
2
This Presentation Include

• Content-Based Image Retrieval System (Brief
  Overview)
• Relevance Feedback (Brief Overview)
• The History of Relevance Feedback
• Issues with Relevance Feedback
• Examples
• My Research Focus

3
What is Content-Based Image Retrieval System?

• Content-based image retrieval (CBIR) are systems
  that retrieved images based on the semantic or
  visual content of the images.
• CBIR systems can be grouped into two main
  categories
• Generic. E.g., QBIC, Lycos, Google and etc.
• Domain Specifics. E.g., medical diagnostic
  system, facial recognition for security system.

4
What is Content-Based Image Retrieval System?
5
The Architecture Layout of CBIR
6
What is Relevance Feedback?

• A strategy that invites interactive inputs from
  the user to refine the query for subsequent
  retrieval.
• An iterative fine tuning query process.
• It act as a bridge between the machine and the
  users.

7
Relevance Feedback In CBIR
8
History of Relevance Feedback In CBIR

• Early Year (Mid 90s - Late 90s)
• Query Point Movement
• Move Closer to the targeted group.
• Re-weight
• Change the searching neighborhood area of the
  query

9
History of Relevance Feedback

• Current State of the Art
• Long term memorization
• Pattern Recognition Problem
• Using Neural Network for SOM
• Classification Problem
• Support Vector Machine
• Machine Learning Problem
• Bayesian Inference Approach
• Integrating Low Level Visual Features with more
  Abstract Conceptual Level.

10
Issues Involved in Relevance Feedback

• Singularity issue in training samples
• Occurs when the feature elements are more than
  the provided training samples.
• Relevancy of the feedback image.
• Pre-clustering Vs Relevance Feedback
• (Efficiency Vs Flexibility)

11
(No Transcript)
12
(No Transcript)
13
(No Transcript)
14
My Research Focus

• Long Term Memorization
• Classification problem
• Feature Integration
• More Specifically how can the machine selects
  feature/s that can be used to best differentiate
  the different image classes with minimal human
  interaction.

15
(No Transcript)