Applying K-means Algorithm for Image Segmentation

This article discusses the application of the K-means algorithm for image segmentation. Unlike traditional segmentation approaches, our method takes color images as input but converts them to grayscale for processing convenience. The implementation involves using pixel intensity values as features for clustering individual pixels. However, illumination variations can cause significant differences in grayscale values among pixels that should belong to the same object, potentially leading to misclassification during the clustering process. To address this issue, the algorithm incorporates a median filtering step at the final stage to eliminate noise points that appear as anomalies on object surfaces in the segmentation results. The K-means implementation typically involves these key steps: converting RGB images to grayscale using weighted channel combination (often 0.299R + 0.587G + 0.114B), initializing cluster centroids, iteratively assigning pixels to nearest clusters based on intensity similarity, and updating centroids until convergence. The median filtering operation uses a sliding window approach to replace each pixel value with the median of its neighbors, effectively smoothing the segmented image while preserving edges. This comprehensive approach enables more accurate segmentation of image components, facilitating better subsequent processing and analysis.