A Content-Based Image Retrieval System Using Custom Convolutional Neural Network for Enhanced Image Retrieval on the Wang Dataset

A large collection of digital images has resulted from the quick growth of multimedia technologies, making effective retrieval a difficult undertaking. Improving retrieval performance, especially in benchmark datasets like the Wang dataset, requires efficient image feature extraction and representation. In this study, a Content-Based Image Retrieval (CBIR) system uses deep learning techniques to enhance the connection between low-level features and high-level semantic representations. Traditional methods require manual selection and extraction of features, which is time-consuming and may not capture relevant information from all the images. The primary objective of this study is to introduce a custom Convolutional Neural Network (CNN) that improves retrieval precision by enhancing feature extraction and addressing the semantic gap. Furthermore, data augmentation is added to make the CBIR system rotation-invariant, which can handle variations in image orientations. In this research work, the Wang dataset is used for experimental evaluation. The system retrieves similar images by comparing features from query images and dataset images based on similarity measurement using Euclidean distance. The performance of the system is evaluated by using Precision at k (P@k), Average Precision at k (AP@k), and mean Average Precision at K (mAP@k) across all classes in the wang dataset. Evaluation results show that the proposed Custom CNN-based CBIR system achieves mAP@20 at 97.8<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup> and P@20 at 96.5%, which is high performance compared to the existing methods. This approach has potential applications in real-world domains like traffic surveillance, face recognition, wildlife conservation, and architectural design, where CBIR plays an important role.