A Method for Solving Image Visual Clustering Problems with Graph Partitioning Approach

This paper introduces a novel methodology for addressing image visual clustering challenges. Unlike traditional approaches focusing on local features and continuity, our method extracts global image impressions by transforming image segmentation into a graph partitioning problem. We propose a global discriminant criterion called Normalized Cut (Ncut), which simultaneously measures inter-cluster dissimilarity and intra-cluster similarity. To optimize Ncut efficiently, we developed an algorithm based on generalized eigenvalue decomposition and applied it successfully to static image segmentation tasks.