Related Experiment Video
Updated: Jul 19, 2026
08:27
Image Recognition and Parameter Analysis of Concrete Vibration State Based on Support Vector Machine
Published on: January 5, 2024
Multitraining support vector machine for image retrieval
Summary
This study introduces multitraining SVM (MTSVM) to improve content-based image retrieval (CBIR) performance with limited labeled data. MTSVM enhances precision and robustness over traditional SVM-based relevance feedback (RF) methods.
Area of Science:
- Computer Science
- Machine Learning
- Image Retrieval
Background:
- Support Vector Machines (SVMs) are common in content-based image retrieval (CBIR) relevance feedback (RF).
- SVM-based RF struggles with small labeled datasets due to classifier instability and high feature dimensions.
- Existing methods show poor performance when training data is limited.
Purpose of the Study:
- To develop a novel machine learning technique to overcome limitations of SVM-based RF with small sample sizes.
- To improve the effectiveness and robustness of relevance feedback in CBIR systems.
- To mitigate issues of SVM instability and ineffective kernel methods in high-dimensional feature spaces.
Main Methods:
- Introduced a new technique: multitraining SVM (MTSVM).
- MTSVM combines cotraining and random sampling in the feature space.
- Algorithm designed to address SVM instability and high feature dimension challenges.
Main Results:
- MTSVM consistently improved performance over conventional SVM-based RFs.
- Enhancements observed in terms of precision and standard deviation.
- Experiments conducted on a large dataset of approximately 20,000 images.
Conclusions:
- The proposed MTSVM algorithm effectively mitigates performance issues in SVM-based RF with limited data.
- MTSVM offers improved effectiveness and robustness for CBIR systems.
- This approach represents a significant advancement for machine learning in image retrieval.
