Abstract

The exponential growth of digital image data necessitates advanced techniques for efficient image mining and clustering. This research proposes a novel approach that integrates semantic mapping with spanning tree optimization to enhance the accuracy and efficiency of image clustering operations. Traditional image clustering methods often struggle with high-dimensional feature spaces and semantic gaps between low-level visual features and high-level semantic concepts. Using deep learning-based feature extraction to build a semantic feature space and minimum spanning tree (MST) optimization for hierarchical clustering, our proposed framework addresses these issues. Convolutional neural networks (CNNs) are used to extract semantic features, graph-based representations are used to represent image relationships, and Prim's algorithm is used to build MSTs. The experimental results on benchmark datasets like CIFAR-10, the ImageNet subset, and custom domain-specific collections show significant advancements over conventional clustering techniques. The proposed approach achieves an average clustering accuracy of 87.3%, a 12.4% improvement over k-means clustering, and reduces computational complexity by 34% compared to hierarchical agglomerative clustering. The semantic mapping layer successfully bridges the gap between visual features and conceptual understanding, while the spanning tree optimization ensures optimal cluster formation with minimal redundancy. This research contributes to the advancement of content-based image retrieval systems, automated image annotation, and large-scale visual data organization.