Image Segmentation Using SVM with K-means Generated Two-Class Maps

This article explores an image segmentation approach using Support Vector Machine (SVM) classification. We begin by generating two-class segmented images through K-means clustering algorithm, which partitions pixels into foreground and background clusters based on color/intensity similarity. The implementation typically uses OpenCV's cv2.kmeans() function with K=2 clusters and appropriate color space conversion. For training data collection, we implement an interactive interface where users select 2*num coordinate points via mouse clicks, stored as 2D position vectors. The first num samples represent positive class regions (target objects) while the subsequent num samples correspond to negative class regions (background). These labeled vectors are then formatted using numpy arrays and fed into Scikit-learn's SVM classifier (sklearn.svm.SVC) with RBF kernel for optimal separation. This methodology enables effective image segmentation by combining unsupervised clustering for initial region proposal with supervised SVM for precise boundary refinement. The approach allows researchers to extract meaningful insights and analytical information from segmented images through systematic sample labeling and machine learning classification.