EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

An Efficient Feature Extraction and Image Retrieval Based on Convolutional Neural Networks

Sawsan Mousa Mahmoud, Hanan A. S. Al-Jubouri

Journal of Studies in Science and Engineering · 30.06.2026 · 6 (1) · 395-415

DOI: 10.53898/josse2026627 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Studies in Science and Engineering, online erschienen am 30.06.2026. Verfasst von Sawsan Mousa Mahmoud und 1 weiteren Autor:innen. Die Arbeit nennt 27 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Image retrieval has become increasingly important in many application domains, including medicine, forensics, biology, astronomy, fashion, and military systems, where large volumes of digital images are generated and stored daily. Consequently, there is a growing need to retrieve images based on their visual content, such as color, texture, and shape, rather than relying on keywords or textual annotations that may be incomplete or unavailable. Content-Based Image Retrieval (CBIR) has been developed to address this challenge. However, designing effective image features remains difficult due to the semantic gap between human perception and machine-level representation. This research proposes a novel hybrid feature extraction and fusion approach for image retrieval, termed Fuse Convolutional Neural Network Models (FCNNM). The proposed method leverages multiple layers of Convolutional Neural Networks (CNNs) to learn discriminative image representations while avoiding the high dimensionality and computational complexity associated with dense layers in traditional Deep CNNs. Instead, the model relies on sequential convolutional and pooling layers. Two CNN architectures with three and four layers, respectively, are designed, and their extracted features are fused to improve retrieval performance. The proposed approach is evaluated using benchmark datasets, namely Corel-1K and Corel-5K. Experimental results demonstrate that fusing 2D CNN 3- and 4-layer models significantly enhances retrieval accuracy, achieving 97% on Corel-1K and 95% on Corel-5K, while maintaining high computational efficiency.

Zitieren

APA

Mahmoud, S. M., & Al-Jubouri, H. A. S. (2026). An Efficient Feature Extraction and Image Retrieval Based on Convolutional Neural Networks. Journal of Studies in Science and Engineering, 6(1), 395-415. https://doi.org/10.53898/josse2026627

Vancouver

Mahmoud SM, Al-Jubouri HAS. An Efficient Feature Extraction and Image Retrieval Based on Convolutional Neural Networks. Journal of Studies in Science and Engineering. 2026;6(1):395-415. doi:10.53898/josse2026627

BibTeX
@article{Mahmoud2026,
  author = {Mahmoud, Sawsan Mousa and Al-Jubouri, Hanan A. S.},
  title = {An Efficient Feature Extraction and Image Retrieval Based on Convolutional Neural Networks},
  journal = {Journal of Studies in Science and Engineering},
  year = {2026},
  volume = {6},
  number = {1},
  pages = {395--415},
  doi = {10.53898/josse2026627}
}
Kostenlos starten →