Smart adaptive learning and optimized feature clustering for enhanced image retrieval

In recent years, Content-Based Image Retrieval (CBIR) has faced increasing challenges due to the rapid growth of multimedia content on the Internet. One of the primary difficulties in CBIR lies in identifying dissimilarities between various objects. This paper introduces a novel approach called SEGJO (Scaling Factor and Elite Opposition Learning-based Golden Jackal Optimization) for effective clustering of extracted features in CBIR. The inclusion of a Scaling Factor (SF) and Elite Opposition Learning (EOL) in SEGJO enhances its search capabilities and helps prevent premature convergence during the clustering process. CBIR extracts key features related to texture, shape, and color using techniques such as Local Binary Pattern, Zernike Moments, and Color Moments. Additionally, an Entropy-based Divergence (ED) function is incorporated into a Convolutional Neural Network (CNN) named EDCNN to improve matching accuracy by reducing redundant activation in hidden layers. The proposed SEGJO-EDCNN method is evaluated on the Corel 5 K and Oxford Flower datasets, and its performance is measured using precision, recall, F-score, mean average precision, and average recall. Comparative analysis with existing methods-ELNDP, SVM-CBIR, SPDNN, and DNN-SAR-demonstrates that SEGJO-EDCNN achieves a higher mean average precision of 97.595% on the Corel 5 K dataset, outperforming both ELNDP and DNN-SAR. Next, the MAP of SEGJO-EDCNN for the Oxford flower dataset is 99.239% which is higher compared to the SVM-CBIR.