Abstract
This study uses the Object-Based Classification OBC method to extract buildings from World-View-3 satellite imagery with high spatial resolution for urban environments. The results are compared to traditional classification Pixel-based methods such as Supervised Classification and Unsupervised Classification. To apply this method, Tarout Island was chosen because of the diversity of its surroundings in terms of water, vegetation and marshes, and the diversity of its urban planning ranging between ancient and modern neighborhoods. Buildings were extracted using Nearest Neighbor and Decision Rules in the OBC method. To test the accuracy, an accuracy assessment was applied. Also, a number of statistical comparisons of results were made with traditional classification Pixel-based methods that included IsoData classifiers, K-means in the Unsupervised Classification, Maximum Likelihood Classifier, Minimum Distance to Means Classifier, and Parallelepiped Classifier in Supervised Classification. Thematic maps were produced using all the previous methods, and their quality was assessed using the Overall Accuracy test, and Kappa Coefficient. The study concluded that the OBC methods achieved high accuracy in extracting buildings in urban areas compared to the traditional Pixel-based methods. The study recommends that more models could be developed to help improve classification in the OBC environment.