Collapsed buildings extraction using morphological profiles and texture statistics - A case study in the 5.12 wenchuan earthquake

The paper proposes a method of collapsed buildings extraction from post-earthquake airborne images. Firstly, training and validation samples are selected, and feature extraction is conducted through morphological profiles and texture statistics based on gray-level co-occurrence matrix, and then support vector machine classifier is used to extract collapsed buildings. An initial experiment is carried out on images acquired after the 5.12 wenchuan earthquake. Results show that morphological features and texture features are complementary to each other in describing collapsed buildings; a reasonable extraction result is achieved based on these two kinds of features. However, due to the complexity of collapsed buildings in airborne images, more work is needed such as introducing more features and also a proper feature reduction method.