The exponential expansion of digital visual material requires sophisticated and efficient picture retrieval technologies. The article presents a new framework using AI to integrate form descriptors, texture patterns, and color histograms to improve Content-Based Image Retrieval. Because they use various feature sets, earlier CBIR systems generally have semantic mismatches and poor retrieval accuracy. We overcome these limits using deep feature fusion with color moments, GLCM for texture analysis, and Fourier and Hu moments for form extraction. SVM classifiers and CNNs are used to add acquired features to a learnt similarity space. This improves semantic matching and retrieval. Attention-guided weighting adjusts feature type importance based on the query image environment. After training and validation on Corel-1 K and Caltech-101 benchmark datasets, the method beats standard CBIR models in accuracy, recall, and mAP. Experimental data shows that the hybrid AI-enhanced CBIR model has a maximum accuracy of 92.3% on Corel-1 K and performs consistently across all categories and noise levels. Improved, more scalable image retrieval systems that combine deep learning-based optimization with low-level visual descriptors will assist in digital asset management, medical imaging databases, and multimedia search engines.
Naveen et al. (Fri,) studied this question.