Abstract: Abstract :ObjectiveTo increase the accuracy of urban building extraction from high-resolution satellite image, this paper proposed anew object-oriented extraction method based on the traditional building extraction approaches, which combined multiple scale segmentation and rules database according to complex urban surface. MethodThe method firstly conducted scale segment using Full Lambda-Schedule algorithm for QuickBird multi-band and full-band merged data. Then, the knowledge rules was established according to the spectral characteristics, shape features, geometric features and texture features and other indicators, which was used for extracting building information from the scale segment result. This article put Baiyun District of Guangzhou City as the study area, and extracted building information utilizing a combination of scale segmentation and rules database.Result and conclusionThe studied results showed that the new object-oriented classification method based on rules can effectively avoid the appearance of phenomenon of salt and pepper in a traditional classification based on pixel, and decrease misclassification (such as roads and shadow), which are more consistent with human way of thinking and closer to the actual value. And, comparing to the traditional pixel-based classification and object-oriented classification, the new classified method produced the highest accuracy with overall accuracy of 87.0154%% and a Kappa of 0.8714, which tested that this new classified method for updating city building thematic database would be more suitable than the general object-oriented classification.