Due to its applications in a variety of fields, including healthcare, surveillance, and e-commerce, content-based image retrieval (CBIR) has recently attracted a lot of attention. By introducing an effective multidomain feature analysis engine that includes incremental learning and continuous feedback operations, this paper introduces a novel method for CBIR. The suggested framework combines a variety of feature extraction methods, such as Fourier, Entropy, Color Map, S, Z, Laplace, GRU, and LSTM, to extract key visual characteristics from images and samples. By effectively maximizing feature variance levels, an Elephant Herding Optimizer (EHO) is used to increase the discriminative power of the chosen features. The automatic selection of the most informative features is made possible by the EHO algorithm, improving retrieval performance levels. The use of a Vector Autoregressive Moving Average (VARMA) model, which successfully captures the temporal dependencies and correlations within the image dataset, further enhances the CBIR process. This model greatly improves retrieval accuracy levels and makes it easier to predict relevant images using the extracted features. Additionally, correlation learning operations are used to incorporate feedback learning, which enables the CBIR system to change and advance over time in response to user feedback. As a result of the system\'s use of the feedback data, the retrieval process has been improved, yielding better precision, recall, and accuracy values for different datasets & samples. These include ImageNet, MIRFLICKR, and CIFAR-10 as well as various real-world image samples were used in the experimental evaluations. The findings show that the suggested approach outperforms current CBIR models with precision of 0.98, recall of 0.985, and recall of 0.97 for different use cases. Further, the proposed system achieves superior accuracy scores and displays lower delay values for real-time scenarios. The increasing demand for accurate and effective image retrieval systems across a variety of applications drives the necessity for this work under real-time scenarios. The proposed framework addresses the limitations of existing CBIR models by combining multiple feature extraction techniques, optimizing feature variance, utilizing temporal dependencies, and incorporating continuous feedback learning process. The potential applications of this research are diverse and include retrieval of medical images for diagnosis and treatment, effective image analysis through surveillance systems, and personalized product recommendations through e-commerce platforms. The proposed approach has several advantages over current CBIR models, including improved retrieval accuracy, adaptability to shifting user preferences, and improved performance levels.