Research On Shadow-based Automatic Extraction Of Buildings From Remote Sensing Images

Recent years, with the improvement of spatial resolution of the satellite sensors, high resolution remote sensing image has become one of the most important data sources for urban information extraction, which is an excellent data source for a large number of applications such as topographic mapping, geo-information systems, environmental modeling and monitoring, urban and regional planning. Man-made object extraction, especially building extraction from remote sensing images has received considerable attention in past decades and is still a hot topic. This thesis has studied the main technologies of automatic building detection from remote sensing images based on building shadow, mainly including shadow detection, tree shadow removal, edge tracing and region growing. The major contributions of the thesis are as follows.1. An automatic urban building extraction method based on shadow is proposed. The method makes good use of shadow cast by buildings in remote sensing images, including shadow detection, tree shadow removal, edge tracing, building's ROI (Region of Interest) selecting and building extraction.2. An automatic shadow detection method for remote sensing images based on gray histogram is proposed. The method fits the gray histogram of the panchromatic remote sensing image by its potential function and finds the first vale point by first difference. Shadow detection is then performed by threshold segmentation using the gray value corresponds to the vale point. The experiment result shows that this method is capable for selecting the segmentation threshold rapidly and automatically in complex images and the shadows are detected correctly.3. A tree detection method based on statistics is proposed, which is mainly composed of color space transform and tree color modeling. The color of trees in the intensity detachable HSI, Lab color space is firstly modeled, then the probability threshold is set properly and the trees are hereby detected exactly by look-up table method in remote sensing images. Experimental results are used to analyze three different models mentioned in the thesis and show that the tree model in Lab color space is most capable of detecting tree regions in high resolution remote sensing images.4. A novel method for tree shadow removal from building shadow is proposed. In high resolution remote sensing images, the detected building shadows are always influenced by the shadows cast by trees. A tree shadow removal method is proposed according to the detected tree shadow and the sun orientation. The method increases the detecting accuracy and is proved effectively by the experiment results.5. Based on the building shadow and building's rectangular model, a method for extraction and description of building region is proposed. The building shadow is firstly processed by edge tracing and vector compressing, and then the gained shadow edge segment list is used to select the building's ROI, which is expanded by region growing algorithm next starting with its center.Based on Matlab R2007a and the remote sensing images from Google Earth, this thesis simulated all the algorithms above and analyzed the results in detail.