Content Based Image Retrieval System Using Deep Learning CNN with ResNet50 Backbone

Abstract

This work proposes a Content Based Image Retrieval structure that leverages the power of deep learning, specifically the Residual Neural Network (ResNet50) architecture. This approach enables accurate and efficient retrieval of visually similar images based on user queries. A pre-trained ResNet50 model is employed to excerpt high level parameters from images. These features capture intricate visual patterns and relationships, providing a robust representation of image content. A curated dataset of 1,000 images is utilized for training and testing the structure. This stored data ensures diversity and covers a wide range of visual content. When a user submits a query image, the model extracts feature from the query. These feature values are compared to the values of images in the storage data using cosine similarity. The top 10 most analogous images are retrieved and displayed to the user. The proposed model offers efficient retrieval, better accuracy and flexibility compared to traditional content based image retrieval systems.