Automatic object-level change interpretation for multispectral remote sensing imagery

In this work, we propose an algorithm for the automatic, qualitative explanation of object-level changes in multispectral remote sensing imagery. This multi-stage approach is based on the construction of change models that directly reflect the nuisance effects of the data acquisition process, unimportant causes of observed object variation, such as illumination variation or spectral composition variation, and important changes such as object shape or location variation. Analysis of the different change models' fit to the data provides a basis for a multi-cause change description of recorded time-variations that objects in the imagery undergo.