IMPROVING IMAGE INTERPRETATION WITH INTEGRATED HYBRID SEGMENTATION AND CLUSTERING APPROACHES

Abstract

Image segmentation is a key aspect of computer vision applications, allowing the division of an image into different regions for analysis. In this study, we introduce a hybrid clustering approach that combines K-Means, Fuzzy C-Means (FCM), and Cluster Grouping Feature-weighted Fuzzy C-Means (CGFFCM) to provide enhanced segmentation accuracy and stability. First, K-Means clustering is applied to initialize cluster centroids, and then the refinement is conducted with FCM to address uncertainty in data. Last, CGFFCM finetunes the cluster assignments by integrating feature weighting and learning cluster variances adaptively. The new approach is compared with the traditional K-Means clustering algorithm to gauge its performance. Performance measures like Accuracy, F-Measure (FM), and Normalized Mutual Information (NMI) are utilized to evaluate the segmentation performance. Experimental results show that the hybrid clustering algorithm outperforms conventional K-Means consistently in segmentation quality, with greater accuracy and improved clustering consistency. This method is especially beneficial in situations where accurate segmentation of intricate images is needed, providing a balance between computational complexity and segmentation performance.