IMAGE RETRIEVAL USING DEEP LEARNING AND GRAPH THEORY
The aim of this research is to retrieve a query image from a pre-defined database of images by extracting features based on similarity measures, taking into account the visual representation of the query image. This includes considerations such as color and orientation, which may vary due to rotation or changes in the viewing angle. In our research, we proposed a specialized Multilayer Perceptron (MLP) trained using Back propagation neural network algorithm designed to function as a visual feature fusion system to enhance the matching process. In addition, we have developed a custom binary matching system in Python to serve as a Content-Based Image Retrieval (CBIR) system, and precision in retrieval of the images. We used Graph Theory, where each image represented as a graph according to color histograms of separated areas. Based on Graph Convolutional Network (GCN), we obtained structural embedding to capture structural features pairing visual features with spatial connections. This type of graph-based algorithm supplements the neural network, and it gives a stronger matching in the images that possess complex structures. The joined system can enhance the overall retrieval effectively with features of pixel-level, relational, and the unification of information features
Authors
- Arjuwan M. A. Al-Jawadi
- Ayhan A. K. Al-Shumam
Publication Details
- Journal
- Kufa Journal of Engineering
- Published
- 2026-11-14
- DOI
- https://doi.org/10.30572/2018/kje/170430
- Primary Topic
- Image Retrieval and Classification Techniques
- Type
- article
- Field-Weighted Citation Impact
- 0.00