Building Extraction from Polarimetric SAR Data using Mean Shift and Conditional Random Fields

Inhalt:
This paper presents a classification framework for extracting buildings from polarimetric SAR (PolSAR) data. Buildings in SAR data are generally composed of layover and shadow regions. First, mean shift bottom-up segmentation approach divides a SAR image into small homogeneous patches. Then conditional random fields (CRF) framework is applied to classify the patches into layover, shadow and other regions. The spatial connectivity between layover and shadow regions is exploited to improve the accuracy of CRF shadow detection. Promising segmentation results of buildings are presented and compared to the results of a basic logistic regression classifier.