| Currently,extracting information such as artificial ground objects and natural scenes from high-resolution remote sensing images has become a hot topic in basic research on remote sensing applications.The research direction of building extraction is particularly hot,and it has attracted more and more attention of scholars.However,due to the influence of image quality,shooting weather conditions,environment and other factors,the current building extraction methods have some limitations,which also make the building extraction results different from the actual building shape.In order to extract building targets from high-resolution remote sensing images with high accuracy and make them reach the practical application level,it is particularly important to improve the regularity and accuracy of building contour.This paper will take high-resolution remote sensing images as the main research data source,and propose a building contour optimization method based on the main direction,aiming at the problems existing in the current building optimization methods,such as sawtooth,low integrity of optimization results and inconsistency with the real building shape.The main work of this paper is as follows:1)This paper first introduces the definition and characteristics of buildings in remote sensing images.By analyzing the main characteristics of buildings,a right-angle polygon building model is introduced and used as a candidate target for building optimization in this paper.Then this paper describes in detail the preprocessing method of the initial extraction results of buildings,which are extracted by the classification method.Firstly,according to the consistency of building area in the same area,the method removes the obvious non-building spots in the initial extraction results.Secondly,we initially simplify the contour and eliminate the disordered line segments and abrupt nodes.The experimental results show that this method can preliminarily eliminate the wrong classification and simplify the building outline.2)Through preliminary simplification,it is found that there are still jagged and irregular problems in the building contour.In view of this,this paper proposes a building contour regularization method based on the main direction.Firstly,this paper introduces the research status of the main direction of buildings in remote sensing images.We describe in detail the steps of extracting the principal direction by combining the bounding rectangle,and then regularize the contour line segments based on the principal direction.Secondly,this paper constructs the complexity model of building contour,converts it into adaptive threshold,and applies it to the regularization algorithm.Finally,we have carried out experiments on empirical threshold and adaptive optimization threshold respectively,which show the effectiveness of automatic optimization of contour regularization.3)In order to solve the problem of edge offset in the regularization results,this paper proposes a method of building contour depth optimization based on the original image and the initial extraction results.This method first calculates the area with high SSIM index based on the original remote sensing image,and then establishes the boundary feature intensity to offset the building contour.The experimental results show that this method effectively improves the accuracy and integrity of the building contour,and makes the contour optimization results more consistent with the real building shape. |
