ICBIRS: Enhanced Content-Based Image Retrieval for High-Precision Image Matching with a Novel Convolutional Neural Network Variant

With the cloud-based ecosystem, various organizations have increasingly adopted the use of multimedia objects, such as images. In many real-world applications, retrieving required images plays a very crucial role. In this context, providing a query to a system is very important, and giving a query by example plays a vital role in retrieving images that satisfy the user's intent to a greater extent. Traditional image processing approaches suffer from the requirement for scalable processing. The emergence of artificial intelligence has enabled learning-based approaches that can serve as improved deep learning models, which are widely used for image processing and require enhancement to realize a content-based image retrieval system. In this paper, we propose a deep learning-based framework known as an Intelligent Content-Based Image Retrieval System (ICBIRS). The system employs an AI-enabled approach for both offline and online phases, extracting features from the database and processing user queries. To extract feature embeddings from a database of images and query images, we proposed and enhanced the CNN model with an attention mechanism and multi-scale feature extraction, enabling efficient retrieval of features and feature embeddings. The proposed system can retrieve top images similar to the query image and reflect user intent as much as possible with even semantic similarity. We proposed an algorithm known as Intelligent Learning based Image Retrieval (ILbIR), intelligent learning-based image Retrieval. The proposed system is evaluated with a benchmark data set. The results revealed that the proposed enhanced CNN model-based approach could leverage image Retrieval performance with the highest accuracy at 97.35%. Therefore, the proposed system can be integrated with real-time multimedia applications where there is a need to retrieve images in a query using an example paradigm.