Enhancing Content-Based Image Retrieval with a Stacked Ensemble of Deep Learning Models

Our research paper delves into an innovative exploration of content-based image retrieval (CBIR), harnessing the capabilities of deep learning models to transform the way images are searched and accessed in extensive databases. The primary goal of our study is to create an ensemble stacking model that synergizes the strengths of various deep learning architectures, thereby boosting the accuracy and efficiency of image retrieval processes. We utilize the Corel Images dataset, rich in diverse visual themes, to test and validate our model’s efficacy. Our research methodology encompasses four key stages: data preprocessing, model training with DenseNet, MobileNet, and Inception-ResNet, followed by an in-depth evaluation of the model. The approach demonstrates the effectiveness of our ensemble model as it achieves a high accurate rate of $\mathbf{9 7 \%}$ exceeds the benchmarks set by the individual models in our compare and differential analyses. Moreover, the manuscript investigates the model’s technical detail, such as the feature extraction, runtime with different images, and scalability to more massive datasets. The eulogy for the model’s performance in the performance evaluation section encapsulates the functional performance and the practicality of CBIR efficacy.