Estimating the extent of damage caused by natural disasters is necessary for implementing effective recovery measures. Damage detection from high-resolution satellite or aerial imagery for post-disaster analysis has been a major research effort in the past decade. A careful analysis of images from before and after an event facilitates rapid detection and assessment of building damage. This work presents a first-of-its-kind system for automatic damage assessment. The proposed framework for damage estimation consists of three steps. First the pre-event and post-event images are registered automatically. A SURF-based feature extraction and matching technique is used for automatic image registration. Next, the objects of interests such as buildings are extracted from pre-storm images. A novel robust algorithm for building detection is proposed and evaluated. Lastly, change detection is performed and damage is classified using supervised learning algorithms. Relevant features that reflect the spectral properties of damaged buildings are identified and used to classify the damage level into various states.