A Study Of Building Extraction From High Resolution Remote Sensing Images

With the rapid development of space technology and sensor technology, Resourcessatellites, reconnaissance satellites, aerial cameras acquire higher and higher ground imageresolution. and you can get all-weather, all-round, real-time observation data. Therefore, theuse of satellite images or aerial photographs for feature information extraction, can greatlyimprove work efficiency, overcome a cumbersome field survey. How to take advantage ofthese high-resolution images are research topics of urban planning, population estimates,resources exploration, military reconnaissance and other areas. In many of these applications,the building is an important part of urban surface features, its acquisition and identification isessential. Therefore, the extraction of the building in high-resolution remote sensing imagesis one of the main research in the field of image processing.This paper focuses on the extraction method in remote sensing images of buildings, thecharacteristics of this paper is first carried out a detailed description, On the base of featureextraction, Considering the relationship between buildings and morphology. this paperpresents a method of the extraction of buildings which are based on a Morphological buildingindex：Firstly we extracted the urban impervious feature, then EMBI was built based on therelationship between the properties of buildings and morphological operators. Subsequently,the EMBI feature image combined with the shape characteristics (length-width ratio, area, etc.)completed the final building extraction using a decision tree method. Experimental resultsshow that the EMBI algorithm extraction algorithm achieves better results with respect to theMBI algorithm.For high-resolution images lack of spectral information, the details of the imageinformation, mixed pixel decrease, pure pixel increases, and the resultant image featureextraction and pattern classification problem, we propose a joint spectrum-spacemulti-feature model based on SVM classifier, using the model to complete the extraction ofthe building, the model takes full advantage of multi-feature information, overcomes the Hughes phenomenon and the accumulation of High-dimensional feature generated over-fittingproblems.The model uses three types of spectral-spatial characteristics, including spectralcharacteristics-spectral characteristics of multi-scale morphology,spectral features multi-scalemorphological features of the underlying surface feature component,spectralfeatures-multi-scale morphological features of extension.firstly using the SVM to classify thespectral-spatial characteristics,then classification results to integrate using the method ofprobabilistic fusion to complete the extraction of the building. Experimental results show thatcompared to the VS-SVM algorithm, the model achieves a better extraction.