Hierarchical urban object extraction from high resolution remotely sensed imagery based on Wavelet Transform and Support Vector Machine

Considering the characteristics of high resolution remote sensing images, we present a new approach based on Wavelet Transform (WT) and Support Vector Machine (SVM) to extract objects from this kind of images. In this paper, the QuickBird image of Xuzhou City, Jiangsu Province is experimented. Wavelet Transform is used to build the image pyramid at first. Support Vector Machine and masking operations are performed to extract objects one by one. The larger object in size is extracted from the transformed image with lower resolution, while the smaller one is extracted from the image of higher resolution. The experimental result shows that this multi-scale approach is suitable to process high resolution remote sensing images and can improve the accuracy and speed of object extraction.