Related Experiment Video
Updated: May 29, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
Object segmentation of database images by dual multiscale morphological reconstructions and retrieval applications
Jiann-Jone Chen1, Chun-Rong Su, W Eric L Grimson
1Department of Electrical Engineering, National Taiwan University of Science and Technology, Taipei 10673, Taiwan. jjchen@mail.ntust.edu.tw
Summary
A novel image segmentation method, SEGON, improves object identification accuracy by 21% and enhances large-scale content-based image retrieval performance by up to 42%. This robust approach is suitable for processing vast image databases.
Area of Science:
- Computer Vision
- Image Processing
- Machine Learning
Background:
- Large-scale image processing requires robust segmentation for target identification.
- Existing methods struggle with diverse content and require improved accuracy.
- Content-based image retrieval (CBIR) performance is limited by segmentation quality.
Purpose of the Study:
- To develop a robust object segmentation method for large-scale image databases.
- To enhance the accuracy of image segmentation using a novel background mesh approach.
- To improve content-based image retrieval performance through enhanced segmentation.
Main Methods:
- Utilized dual multi-scale gray-level morphological operations (SEGON) to create a background variation mesh.
- Employed normalized probability random index (PRI) to evaluate segmentation accuracy against hand-labeled images.
- Integrated SEGON into CBIR systems, comparing precision-recall (PR) and rank performances with and without the method.
- Applied AdaBoost for salient shape feature selection and histogram intersection for scalable HSV color descriptors.
Main Results:
- SEGON improved object segmentation accuracy by 21% compared to existing methods, as measured by PRI.
- SEGON-enabled CBIR demonstrated up to 42% improvement in PR performance at a 0.5 recall rate on large databases.
- The method proved robust in handling large-scale image databases with diverse content.
Conclusions:
- The proposed SEGON method offers superior object segmentation accuracy and robustness for large-scale image processing.
- SEGON significantly enhances the performance of content-based image retrieval systems.
- The methodology is extensible for incorporating additional image features to further boost retrieval performance.
