Development of automated extraction method for building damage area based on maximum likelihood classifier

Using some images taken from a helicopter after recent earthquakes, an automatic extraction method of the class of severely damaged buildings was examined by the maximum like- lihood classifier with a multivariate normal distribution for the statistics of some training data in images. In the image processing, hue, saturation, brightness, edge intensity and variance of edge in- tensity were used for the extraction of information on damaged buildings, and the threshold value for each image did not have to be decided empirically such as in the method of Hasegawa et al. (2000b). The classification result was good for most of the classes used by this method. In particu- lar, the class denoting collapsed buildings agreed well with the actual situation of building damage. Therefore, if we could set up some training data in each class, we would obtain favorite results for areas with building damage rapidly, and the application of this method to real-time earthquake dis- aster management can be expected.