Related Experiment Video
Updated: May 21, 2025
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
448
Deep image features sensing with multilevel fusion for complex convolution neural networks & cross domain benchmarks
Aiza Shabir1,2, Khawaja Tehseen Ahmed2, Arif Mahmood3
1Institute of Computer Science and Information Technology, The Women University Multan, Multan, Pakistan.
Plos One
|March 18, 2025
Summary
This study introduces a novel image retrieval method using feature vectors and deep learning. The approach enhances image retrieval accuracy and efficiency across diverse datasets.
Area of Science:
- Computer Science
- Artificial Intelligence
- Image Processing
Background:
- Content-Based Image Retrieval (CBIR) relies on primitive image signatures and feature vectors for classification.
- Efficient retrieval from diverse datasets is essential in the digital age.
Purpose of the Study:
- To develop a unique feature identification technique for improved image retrieval.
- To enhance the discriminating strength of retrieval systems by extracting deep image features.
Main Methods:
- A novel feature identification technique based on suppression and pixel derivative sums to locate interest points.
- Scale space interpolation combining color, shape, and object information.
- Object-based feature vectors using high variance coefficients, converted to bag-of-visual-words (BoVW).
- Multilayer fusion of Convolutional Neural Networks (CNNs) for deep feature extraction (primitive, spatial, overlayed).
Main Results:
- Extensive experiments on standard datasets (ALOT, Cifar-10, Corel-10k, Tropical Fruits, Zubud) show significant improvements.
- Demonstrated considerable enhancements in precision, recall, average precision/recall, and mean average precision/recall.
Conclusions:
- The integration of traditional feature extraction with multilevel CNNs advances image sensing and retrieval.
- The proposed method offers more accurate and efficient image retrieval solutions.
