Deep Representation-Based Fuzzy Graph Model for Content-Based Image Retrieval

Image retrieval involves searching for images relevant to a user-provided query image. In this paper, we aim to develop a graph-based model with deep representations for Content-Based Image Retrieval (CBIR). Inspired by recent advancements in deep learning, we initially employ a fine-tuned Convolutional Neural Network (CNN) to capture deep semantic features for a specific target image database. Utilizing these learned features, we then introduce an graph-based ranking method for online retrieval. This model’s constructed graph is designed to characterize the geometrical structure of the data manifold, facilitating an efficient ranking process. Finally, based on user-provided feedback regarding relevant and irrelevant images, we update the retrieval system in both the deep learning framework and the graph-based ranking model in an offline manner. Extensive simulations confirm the efficiency and effectiveness of our proposed model. © The Author(s) under exclusive licence to Taiwan Fuzzy Systems Association 2024.