Quasi-Oppositional Satin Bowerbird with Deep Learning based Content based Image Retrieval

Abstract

Content-based image retrieval (CBIR) is commonly employed to retrieve images from a massive set of unlabeled images. The design of CBIR model faces several limitations, as it is mainly based on the extraction of image features to calculate the similarity amongst the query image (QI) and database images. The recent advances of deep learning (DL) models help to attain remarkable retrieval outcomes. In this view, this paper presents a novel quasi-oppositional satin bowerbird optimizer with Densely Connected Networks (QOSBO-DCN) for CBIR. The proposed QOSBO-DCN technique aims to properly retrieve the images related to the QI in an effective and automated manner. The proposed QOSBO-DCN technique derives a DenseNet-77 model as a feature extractor to derive feature vectors from the QI and database images. Besides, the QOSBO algorithm is utilized to adjust the hyperparameter values of the DenseNet-77 model in such a way that the retrieval performance can be improved. Additionally, Euclidean distance is used as a similarity measurement approach to determine the highly resembling images and retrieve them. The simulation analysis of the QOSBO-DCN technique is performed using Corel10K dataset and the results reported the betterment of the QOSBO-DCN technique over the existing techniques.