Abstract: Aiming at the problem that only using pixel-level or featurclevcl change detection for high-rise building has low accuracy, a method combined pixel-level and feature-level change detection was presented. Detect changes of multi temporal remote sensing image based on ratio method, obtain the candidate change regions of high-rise buildings, then detect changes in the candidate regions based on building feature. Firstly,a novel fast registration algorithm of con- straint based Delaunay triangulation is used to make registration of multi-spectral images with two different phase. The building will lead to changes of distribution and characteristics of color in the texture of the local area changes, so, ro- bust texture and color characteristics for radiation and registration are extracted to make change detection. Experiment results show that the combination of pixel-level and featurclevcl of the building change detection method can effectively improve the accuracy, and reduce the false alarm.

Key words: Building, Change detection, Evaluation index, Multi-spectral images