EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

ADVANCED FEATURE REPRESENTATION IN CONTENT-BASED IMAGE RETRIEVAL UTILIZING DEEP LEARNING FRAMEWORKS

Muhammad Mohsin, Jahan Khan, Muhammad Faisal, Asim Abdul Qadir, Muhammad Arslan, Sohail Raza Chohan

Kashf Journal of Multidisciplinary Research · 28.03.2026 · 3 (03) · 149-173

DOI: 10.71146/kjmr880 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Kashf Journal of Multidisciplinary Research, online erschienen am 28.03.2026. Verfasst von Muhammad Mohsin und 5 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Content-Based Image Retrieval (CBIR) systems aid in the retrieval of images by utilizing the inherent visual attributes as opposed to the manual textual labels, providing a better alternative to the conventional metadata-based search. With the growing rate of development of digital repositories in key domains such as medical imaging, surveillance, and digital forensics, the need to undertake automated, scalable and precise image analysis has been stressed. But modern CBIR systems are often faced with bottlenecks including poor feature extraction, ineffective preprocessing pipelines, and insensitivity to environmental changes such as changing lighting and background noise. To overcome these shortcomings, this research will present a new improved CBIR architecture with streamlined image-processing pipeline. The suggested methodology will shift the input of colors to the regular grayscale form since the calculated scale is normalized to reduce the computational burden without affecting the integrity of the structure. A strict preprocessing pipeline including Contrast Limited Adaptive Histogram Equalization (CLAHE), spatial normalization and data augmentation introduced to guarantee feature consistency and resistant Ness to noise. A Convolutional Neural Network (CNN) provides the feature representation, to be more precise the state of the art MobileNetV2 design is used, which employs depth wise separable convolutions to extract features high-effectively. Extraction of feature vectors is handled through an effective indexing scheme to do fast similarity matching. Experimental evidence shows that the proposed system is much better than the currently available baseline approaches in retrieval accuracy (98.06/97.04) and computational efficiency. The results affirm that the system is robust and can be deployed in real-life and high-volume image retrieval systems.

Zitieren

APA

Mohsin, M., Khan, J., Faisal, M., Qadir, A. A., Arslan, M., & Chohan, S. R. (2026). ADVANCED FEATURE REPRESENTATION IN CONTENT-BASED IMAGE RETRIEVAL UTILIZING DEEP LEARNING FRAMEWORKS. Kashf Journal of Multidisciplinary Research, 3(03), 149-173. https://doi.org/10.71146/kjmr880

Vancouver

Mohsin M, Khan J, Faisal M, Qadir AA, Arslan M, Chohan SR. ADVANCED FEATURE REPRESENTATION IN CONTENT-BASED IMAGE RETRIEVAL UTILIZING DEEP LEARNING FRAMEWORKS. Kashf Journal of Multidisciplinary Research. 2026;3(03):149-173. doi:10.71146/kjmr880

BibTeX
@article{Mohsin2026,
  author = {Mohsin, Muhammad and Khan, Jahan and Faisal, Muhammad and Qadir, Asim Abdul and Arslan, Muhammad and Chohan, Sohail Raza},
  title = {ADVANCED FEATURE REPRESENTATION IN CONTENT-BASED IMAGE RETRIEVAL UTILIZING DEEP LEARNING FRAMEWORKS},
  journal = {Kashf Journal of Multidisciplinary Research},
  year = {2026},
  volume = {3},
  number = {03},
  pages = {149--173},
  doi = {10.71146/kjmr880}
}
Kostenlos starten →