Research On Building Extraction Based On High Resolution Color Remote Sensing Image

Recently, the real-time, high-precision and reliable data got from the remote sensing images have been an important source to detect the information of buildings. Because of the rapid development of the remote sensing and computer technology, the resolution ratio of the data information increased significantly and exponentially. To extract the buildings which were regarded as one of import objects in the remote sensing image rapidly make an important practical significance to many fields such as environment monitoring, geographic information updating, agriculture, the disaster prevention and relief. For the moment, the study of building extracting is still an interesting topic. Aim at the shortcomings of building extracting which with shadow, the removal of vegetation, the ground objects shadow detection, resegmentation and following processing had been studied and explored in this article. The main work in this article includes the following several aspects:(1) A vegetation extraction method was put forward after the offline study, and as the prior knowledge the vegetation likelihood model was obtained, for the serious influence of vegetation information when extract the building shadow. Which can identify the experiment image area and removed vegetation, it provides favorable conditions for the extraction of subsequent shadow.(2) The mean shift algorithm was used to segment the image and the image was divided into many small homogeneous areas after studying the algorithm and principle of mean shift.(3) On the premise of over-segmentation, the space transformation between RGB and HSI was realized. Based on the property of lower light and higher color of shadow in the remote sensing image, the ratio of color and brightness was worked out to act as the threshold to realize the shadow area extracting with Ostu threshold method. And then, the morphological post-processing of extraction result was made, the small area tree shadow was filtered, and the building shadow was extracted finally.(4) The nearest shadow area was found by calculating the pixel distance in the shadow expansion area. The candidate areas were determined finally based on the lighting direction and the spatial relation between ground objects and the shadow. And then, the QDA classify was given out by the ratio of red and green, at last, the buildings were extracted out.The experimental data in this article got from the Quick Bird high resolution remote sensing image, the software which could extract the buildings successfully were Visual Studio 2008.net and Matlab. The results will be compared with the original image and detailed analysis, the experiments showed that this method could get some results which may be more complete and more tally with the actual situation.