Related Experiment Video
Updated: Aug 9, 2026
12:27
Large-scale Reconstructions and Independent, Unbiased Clustering Based on Morphological Metrics to Classify Neurons in Selective Populations
Published on: February 15, 2017
CLUE: cluster-based retrieval of images by unsupervised learning
Yixin Chen1, James Z Wang, Robert Krovetz
1Department of Computer Science, University of New Orleans, New Orleans, LA 70148, USA. yixin@cs.uno.edu
Summary
This study introduces Cluster-based Retrieval of Images by Unsupervised Learning (CLUE), a novel technique that improves image retrieval systems by analyzing similarities between images. CLUE enhances user interaction by dynamically clustering images near the query, outperforming traditional methods.
Area of Science:
- Computer Science
- Information Retrieval
- Machine Learning
Background:
- Traditional content-based image retrieval (CBIR) systems primarily focus on query-target image similarity.
- Similarities among target images themselves are often overlooked, limiting retrieval efficiency and user interaction.
Purpose of the Study:
- To introduce Cluster-based Retrieval of Images by Unsupervised Learning (CLUE) to enhance CBIR systems.
- To improve user interaction by leveraging similarity information among target images.
- To demonstrate CLUE's effectiveness in dynamic, query-dependent image clustering.
Main Methods:
- CLUE employs a graph-theoretic clustering algorithm on images proximate to the query.
- Clustering is dynamic, adapting to the specific images retrieved in response to a query.
- CLUE is compatible with various real-valued symmetric similarity measures and can integrate with relevance feedback systems.
Main Results:
- An experimental CBIR system using CLUE demonstrated improved performance on a 60,000-image COREL database.
- CLUE-enhanced retrieval outperformed a standard CBIR system using the same similarity measure.
- The technique showed potential for real-world application with Google Image Search data.
Conclusions:
- CLUE offers a significant improvement over traditional CBIR methods by utilizing inter-image similarity.
- The dynamic clustering approach enhances user interaction and retrieval effectiveness.
- CLUE is a versatile technique adaptable to existing CBIR systems and keyword-based search interfaces.
