Building Extraction Research Based On Clustering Segmentation Algorithm

In the field of remote sensing image processing and computer vision,High-resolution remote sensing imaging technology is becoming an increasingly important research topic in the field of automatic detection and image analysis of geospatial targets.It is of great significance to extract high-precision and reliable buildings and other artificial features from high-resolution remote sensing images.Aiming at some problems existing in buildings extraction,a new K-means segmentation algorithm is proposed based on watershed segmentation in this thesis.Then the polygonal shape of the extracted homogeneous region is fitted.Finally,the contours of the building are regularized and discriminated.The main work in this article includes the following aspects:(1)Image preprocessing:Firstly,in order to reduce the interference of background information,vegetation areas are recognized and removed from the test scene;Secondly,the regions with similar gray scale are divided into block-shaped regions by watershed segmentation,which provides the initial clustering center position and initial K value for the subsequent K-means clustering segmentation,and avoids the undesirable segmentation situation that caused by traditional K-means algorithm selecting improper initial clustering center position and initial K value randomly.(2)New K-means segmentation algorithm proposed in this thesis can effectively solve the difficult problem about selection of initial center position and K value of K-means algorithm,which can be more effective for image segmentation.Then the edge detection algorithm is used to detect the edge information of the segmented closed area and prepare for the subsequent building extraction.(3)Building object extraction:We assume that a closed regularized area indicates one building.On the basis of the edge contours extraction,a polygon fitting algorithm which is based on the dominant points detecting and moment estimating is proposed to optimize the fitting polygon,combined with basing on ISOCLUS clustering of the main direction to estimate the main direction of polygon.Then,according to the first principal direction weight,the outline of the building is normalized to a right-angled or collimated polygon.If the cost of regularization is too large,which means the polygon does not corresponding to a building,along with the regularization be canceled,and the initial polygon will be kept.