Automatic Change Detection in Urban Areas under a Scale-space, Object-oriented Classification Framework

In this paper an object-based classification framework is introduced for the automated monitoring of changes in urban areas. Morphological scale space filtering is embedded in the processing procedure constraining qualitatively the multi-level segmentation and thus, the structure of object hierarchy. The elegantly simplified images provide a more compact and reliable source in order to generate image objects in various scales. Multivariate alteration detection (MAD) transformation is applied on the simplified images towards the identification of changes. Experimental results indicated that important information, regarding the changes of man-made features, is concentrated to the higher order MAD components. While, the first component has, in most cases, maximum variance and ideally carries the maximum amount of information on changes, the second one has maximum spread in its intensities subject to the condition that it is statistically uncorrelated with the first one, and so on. While the man-made changes are unrelated with the changes of phenological cycle and image noise, it is quite common that such changes are highlighted in higher order components. Last but not least, the labeling of the most significant changes is addressed through an effective object-based set of rules over the MAD components. The quantitative and qualitative evaluation of the developed scale-space, object-oriented classification framework indicates the potentials of the developed approach.