Machine learning algorithms for color image segmentation

Abstract

This report presents a complete framework to segment color pictures into regions, classify these regions as "interesting" or "not interesting" and extract color information from the interesting regions. It proposes the segmentation using k-means and the classification using kNN and SVM (Support Vector Machines). It also details the color space transformation and why they need to be done. There is a detailed example of the framework being used in a real dataset with the respective results commented and explained.