Improve Content-based Image Retrieval using Deep learning model

Abstract

The complexity of multimedia has expanded dramatically as a result of recent technology breakthroughs, and retrieval of similar multimedia material remains an ongoing research topic. Content-based image retrieval (CBIR) systems search huge databases for pictures that are related to the query image (QI). Existing CBIR algorithms extract just a subset of feature sets, limiting retrieval efficacy. The sorting of photos with a high degree of visual similarity is a necessary step in any image retrieval technique. Because a single feature is not resilient to image datasets modifications, feature combining, also known as feature fusion, is employed in CBIR to increase performance. This work describes a CBIR system in which combining DarkNet-19 and DarkNet-53 information to retrieve images. Experiments on the Wang (Corel 1K) database reveal a considerable improvement in precision over state-of-the-art classic techniques as well as Deep Convolutional Neural Network(DCNN).