面向对象轮廓约束GGVF Snake的建筑物边界优化方法

分析高分辨率遥感影像中建筑物的特征和常用方法提取建筑物存在边界漏检误检导致的边界不规则等问题，提出面向对象轮廓约束广义梯度矢量流（GGVF）Snake模型的建筑物边界优化方法. 在利用分类法获取建筑物轮廓的初始结果基础上，自动提取每个建筑物轮廓线作为GGVF Snake的初始轮廓线，获取各轮廓外接矩形进行对象裁剪，提取每个建筑物的子图对象. 对每个子图对象进行Canny边缘检测，结合Hough变换提取直线特征，输入到广义梯度矢量流模型的迭代求解中快速最小化能量函数，实现对象级建筑物轮廓的精确优化. 实验结果表明，利用提出的方法能够自动获取初始建筑物的轮廓信息，提高优化的自动化程度；基于对象级的边缘检测与直线特征的加入，有助于GGVF Snake快速拟合，准确地平滑建筑物边缘且准确度高. 相对于其他对比方法，本文方法的轮廓优化总体精度和综合值均有提升，可以作为有效提升分类原理获取的建筑物轮廓的优化后处理手段，提高了建筑物提取的精度.

Building boundary optimization method based on object-oriented contour constraint GGVF Snake model

An object-oriented contour constrained generalized gradient vector flow (GGVF) Snake model for building boundary optimization method was proposed by analyzing the features of buildings and the problems of conventional methods that the irregular boundary problem caused by false classification is pervasive. Each building boundary was automatically extracted as the initial boundary of GGVF Snake based on the initial building results obtained by classification. The circumscribed rectangle of each building boundary was obtained for object clipping in order to extract each building sub-image object. Canny edge detection was performed on each sub-image object to acquire each building boundary result that will be combined with Hough transform to extract linear features. Then the linear features were inputted into the generalized gradient vector flow model’s iterative solution to minimize the energy function. Then the precise of the object-level building contour was enhanced. The experimental results show that the proposed method can automatically obtain the initial building boundary information and improve the automation degree of optimization. The object-level contour detection and the addition of line features help GGVF Snake quickly fit and accurately smooth the building boundary to enhance the contour accuracy. The overall accuracy and comprehensive value of the proposed method are improved compared with other contour optimization methods. Thus the method can be used as an effective post-processing optimization method for the classification principle to improve the building extraction precision.