Abstract—An essential criterion in content-based image retrieval is its accuracy. High accuracy depends heavily on the quality of feature extraction and dimension reduction methods. To achieve this aim, this paper integrates Visual Geometry Group (VGG) 16-based feature extraction with Autoencoder-based dimensionality reduction, along with Grey Wolf Optimization (GWO), to enhance the K-means clustering algorithm. Initially, 4096 deep features were extracted from images using VGG16. These features were then reduced to 10, 15, and 30 latent features using an Autoencoder (AE) that captures complex data structures more effectively than linear approaches. Moreover, the clustering stage was further improved by applying GWO to promote the initialization of K-means centroids. The proposed method was assessed on two datasets: Corel-1k (10 classes) and UC Merced Land Use (UCM) (21 classes). Further, it was compared with the Principal Component Analysis (PCA)-based method and other recent deep learning approaches. The experimental results revealed that the proposed method outperformed other studies in both image retrieval accuracy by mean average precision, and clustering quality, assessed by Purity, Adjusted Rand Index (ARI), and Normalized Mutual Information (NMI). These results confirm that feature extraction using nonlinear dimension reduction by AE yields more discriminative features than PCA and other approaches in all metrics.
 
Keywords—Content-Based Image Retrieval (CBIR), deep feature extraction, Visual Geometry Group (VGG) 16, autoencoder-based dimensionality reduction, grey wolf optimizer, K-means clustering, deep learning
 
Copyright © 2026 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).