Object-Oriented Building Extraction from High-Resolution Imagery Based on Fuzzy SVM

This paper proposes a novel object-oriented building extraction method based on fuzzy support vector machines (SVM). The choice to adopt an SVM classification technique is motivated by the high number of parameters derived from the feature-extraction phase, which requires a classifier suitable to the analysis of hyper-dimensional features spaces. This method can be divided into two different phases: first, extraction of multi spectrum information, texture information, and spatial information about building structures from QuickBird imagery of building; then, integrates all features to classify the buildings. The remote sensing classification using support machine has obtained satisfactory results, but mixed samples often reduce the performance. In this paper, we propose an approach based on nonlinear fuzzy support vector machine, in which the fuzzy membership is calculated by samples´ purification. The results show that this method can significantly improve the accuracy of the building extraction.