An enhanced CBIR System Using Modified VGG16 with Stacked SVM, Random Forest, and XGBoost Classifiers

CBIR systems have improved large image dataset search and management, yet traditional techniques with HSV GLCM and SIFT experience ‎limited precision because of their insufficient feature extraction abilities. The research requires additional efforts to boost retrieval precision and ‎the overall computer vision model functioning, specifically for CNN-SVM constructions. Through VGG16-based feature extraction, a newly ‎proposed hybrid Content-Based Image Retrieval framework incorporates a stacked classification system built using SVM alongside RF and ‎XGBoost for decision-making. Tests on the Wang dataset validated this model by showing the CNN-SVM baseline model delivering precision ‎levels of 83.61% at 10 retrievals, and both 83.67% at 15 and 83.37% at 20 retrievals. Implementing stacked classifiers produced advanced ‎decision boundaries that improved classification performance while raising total precision by 12.06% at all retrieval settings above the baseline ‎model. The research work establishes the basic groundwork for next-generation hybrid CBIR approaches while demonstrating the benefits of ‎deep learning and ensemble technologies in improving retrieval performance.