A Novel Hyperparameter Tuned Deep Learning Model for Content based Image Retrieval

• Published in: 2022 International Conference on Sustainable Computing and Data Communication Systems (ICSCDS) pp. 207 - 213
• Main Authors: Kalaivani, K., Govindaraju, Rohini, Sheeba, Paul T, Sivakumar, P, Pundir, Aditya Kumar Singh, Maragatharajan, M., Jegajothi, B.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 07.04.2022
• Subjects: CBIR (Content Based Image Retrieval); Computational modeling; Computer architecture; Deep learning; DL (Deep Learning); Feature extraction; Image retrieval; Metaheuristics; Parameter Tuning process; Particle measurements; Precision; Real-time systems; Recall
• DOI: 10.1109/ICSCDS53736.2022.9760906

Content based image retrieval (CBIR) becomes useful in several real time applications and the size of the image databases gets significantly improved. At the same time, it is hard to undergo a comparison of the feature of the query image (QI) with all images from the dataset. Due to the advancements of deep learning (DL) models, the CBIR models can be designed to accomplish maximum retrieval performance. With this motivation, this paper presents a novel hyperparameter tuned DL model for CBIR (HPTDL-CBIR) models to effectively retrieve the images in the image database. The presented HPTDL-CBIR model involves a VGG-16 model as a feature extractor for deriving a meaningful collection of feature vectors for the QI and the images in the database. At the same time, water wave optimization (WWO) techniques are applied to properly determine the hyperparameter of the VGG-16 model in such a way that the retrieval performance gets enhanced. Finally, Manhattan distance metric is applied in the query matching procedure to determine the similarity level among the images. In order to validate the better retrieval outcome of the presented HPTDL-CBIR method, a series of simulation takes place. The experimental results showcased that the HPTDL-CBIR method has gained maximum outcome.