Translator Disclaimer
23 August 2017 Hybrid method for building extraction in vegetation-rich urban areas from very high-resolution satellite imagery
Author Affiliations +
A continuous update of building information is necessary in today’s urban planning. Digital images acquired by remote sensing platforms at appropriate spatial and temporal resolutions provide an excellent data source to achieve this. In particular, high-resolution satellite images are often used to retrieve objects such as rooftops using feature extraction. However, high-resolution images acquired over built-up areas are associated with noises such as shadows that reduce the accuracy of feature extraction. Feature extraction heavily relies on the reflectance purity of objects, which is difficult to perfect in complex urban landscapes. An attempt was made to increase the reflectance purity of building rooftops affected by shadows. In addition to the multispectral (MS) image, derivatives thereof namely, normalized difference vegetation index and principle component (PC) images were incorporated in generating the probability image. This hybrid probability image generation ensured that the effect of shadows on rooftop extraction, particularly on light-colored roofs, is largely eliminated. The PC image was also used for image segmentation, which further increased the accuracy compared to segmentation performed on an MS image. Results show that the presented method can achieve higher rooftop extraction accuracy (70.4%) in vegetation-rich urban areas compared to traditional methods.
© 2017 Society of Photo-Optical Instrumentation Engineers (SPIE) 1931-3195/2017/$25.00 © 2017 SPIE
Ajith S. Jayasekare, Rohan Wickramasuriya, Mohammad-Reza Namazi-Rad, Pascal Perez, and Gaurav Singh "Hybrid method for building extraction in vegetation-rich urban areas from very high-resolution satellite imagery," Journal of Applied Remote Sensing 11(3), 036017 (23 August 2017).
Received: 14 February 2017; Accepted: 28 July 2017; Published: 23 August 2017

Back to Top