Abstract: An object-oriented typical ground objects extraction method based on multi-level rules is presented to solve the problem of ground objects extraction,which are complex and various in high resolution remote sensing image.Image segmentation method based on region is improved with an initial growing by global optimal principle,avoiding seed selecting.Image objects from image segmentation are used as primitive for extraction of ground objects extraction.The typical extraction features are analyzed for different ground objects.Multi-level rules are made based on different features and used to extract ground objects.And overall exaction accuracy of 87.1% is achieved in the experiment.