Indian Journal of Science and Technology
Year: 2023, Volume: 16, Issue: 34, Pages: 2693-2702
Ovais Rashid Khan1*, Javaid Iqbal Bhat1
1Department of Computer Science, Islamic University of Science and Technology, Kashmir, India
Email: [email protected]
Received Date:02 June 2023, Accepted Date:02 August 2023, Published Date:08 September 2023
Objective: The objective of this study is to conduct a comprehensive review of existing research and literature in the field of Content-Based Image Retrieval (CBIR). This review highlights the key challenges associated with the extraction and representation of visual semantics of images. This paper discusses the measure used computing similarity and ranking of retrieved images by CBIR system. The review discusses limitation of traditional approaches and also highlights the challenges with the current deep learning methods in semantic feature representation, defining the similarity metrics and indexing. This paper also highlights scalability and generalization challenges in implementing real environment. Methods: A thorough literature review was conducted on wellestablished databases, including Scopus, Web of Science, IEEE Xplore, ACM, and Science Direct, employing appropriate keywords. Mention the period of coverage. Pertinent search terms encompassed local feature representation, global feature representation, low-level features, high level features, semantic gap, image embeddings, handcrafted features, deep learning, image descriptors, similarity, and image indexing, with the aim of exploring content-based image retrieval systems. Comparative analysis was performed on the chosen articles, taking into account factors such as algorithms, methodologies, datasets, and evaluation metrics. The results discussed using comparative analysis, ensuring a comprehensive overview of recent literature on content-based image retrieval, offering valuable insights and highlighting emerging trends in the field. Findings : The research uncovers the novelty in the realm of contentbased image retrieval (CBIR) by highlighting the challenge of high-level visual semantics when comparing images, as perceived by humans. It emphasizes that feature extraction methods and choices significantly influence CBIR system performance, stressing the importance of selecting suitable features and similarity measures based on image dataset characteristics and application requirements. The study underscores the persistent obstacle of the semantic gap between low-level visual features and high-level semantic concepts, encouraging exploration of diverse approaches like deep learning, relevance feedback, and ontology-based methods to bridge this gap. Particularly, deep learn-ing techniques, notably Convolutional Neural Networks (CNNs), have shown promising results in CBIR by automatically learning hierarchical representations capturing high-level semantic information. However, the review also highlights the challenges of scaling deep learning methods and the limited accessibility of precisely labelled datasets, which can hinder performance and generalization across diverse image datasets and real-world scenarios. Deep learning models pose interpretability challenges due to their complex, opaque nature and hierarchical semantic representations.
Keywords: CBIR; ContentBased Image Retrieval; Deep Learning; Convolutional Neural Network; Local Features; Global Features; Similarity Metric; Semantic Gap
© 2023 Khan & Bhat. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Published By Indian Society for Education and Environment (iSee)
Subscribe now for latest articles and news.