Related Experiment Videos
An adaptable image retrieval system with relevance feedback using kernel machines and selective sampling
Mahmood R Azimi-Sadjadi1, Jaime Salazar, Saravanakumar Srinivasan
1Department of Electrical and Computer Engineering, Colorado State University, Fort Collins, CO 80523, USA. azimi@engr.colostate.edu
Summary
This adaptable content-based image retrieval system uses machine learning and Fisher information for accurate underwater object identification. It improves search accuracy by learning from user feedback and classification models.
Area of Science:
- Computer Science
- Artificial Intelligence
- Machine Learning
Background:
- Content-based image retrieval (CBIR) systems are crucial for searching large image databases.
- Existing CBIR systems often struggle with adapting to user-specific needs and complex image data.
- The need for adaptable retrieval systems that can precisely match user concepts or classification models is evident.
Purpose of the Study:
- To present an adaptable content-based image retrieval (CBIR) system.
- To improve retrieval accuracy by incorporating regularization theory, kernel-based machines, and Fisher information measure.
- To enable the system to adapt to either a multiclass classification model or user-defined high-level concepts.
Main Methods:
- Developed an adaptable CBIR system with a retrieval subsystem, multiple adaptive mapping subsystems, and a relevance feedback mechanism.
- Employed regularization theory and kernel-based machines for similarity matching and adaptation.
- Introduced a novel Fisher information-based method for selecting informative query images during relevance feedback learning.
Main Results:
- The adaptation process successfully minimized retrieval error, achieving accurate matching with classification models or user concepts.
- The Fisher information measure facilitated the selection of optimal query images, enhancing relevance feedback learning.
- Thorough testing on an underwater object database demonstrated the system's effectiveness.
Conclusions:
- The proposed adaptable CBIR system effectively enhances image retrieval accuracy through adaptive learning mechanisms.
- The integration of Fisher information provides a robust method for query image selection in relevance feedback.
- The system shows significant promise for domain-specific image retrieval applications, particularly with complex datasets.