Probabilistic Representation of Structural Integrity of Urban Buildings in Remotely Sensed Images

We present a method to represent structural integrity of urban buildings using remotely sensed satellite images. The method involves extracting structural damage indices of individual urban buildings by comparing pre-and post-disaster satellite images, considering image distortions irrelevant to damage. To accomplish this, a probabilistic approach using mixture-of-Gaussians models (MoG), which is based on the extraction of affine-invariant features of structural integrity, is proposed. The Kullback-Leibler (KL) divergence is used to measure the change of structural integrity (degree of structural damage). The effectiveness of this method is demonstrated by conducting damage segmentation and damage index extraction for a single urban building and a group of buildings, respectively, which are destructed by a natural disaster.