针对高分辨率遥感影像中分类法提取的建筑物轮廓不规则问题,设计了一种逐级优化规整建筑物轮廓的方法。根据分类验证思想提取的建筑物的初始结果,首先提取建筑物初始轮廓进行多边形拟合,获取与建筑物轴线倾斜程度相一致的最佳拟合外接矩形,然后将建筑物轮廓线段与最佳拟合外接矩形边界进行等分并比较两者之间的单向Hausdorff距离,对距离较小且满足替换条件的轮廓线段等分点利用对应的最佳拟合外接矩形边界等分点进行替换,实现轮廓初步的规整优化;接着利用Shi--Tomasi算法对建筑物局部无法规整的复杂轮廓区域进行特征角点提取、匹配、排序与剔除,进一步对边缘特征点进行有序连接与重构,实现轮廓深度的优化,最终提高边缘表达准确度和提取精度。通过对多幅遥感影像进行实验对比分析,结果表明本文方法不仅适用于不同分类方法提取的建筑物结果的轮廓优化,有效提高建筑物轮廓的边缘表达精度,而且相比于轮廓优化参照方法,逐级优化能更准确地适应复杂建筑物轮廓的细节变化,优化的精度整体较优,使建筑物边缘的准确性、规整程度得到有效改善,能更真实准确地反映建筑物的真实形状。

This study proposed a method for optimizing and regularizing building contours step by step to investigate the irregular problem of building contours extracted by classification in a high-resolution remote sensing image. The initial building contours were extracted and reconstructed by polygon fitting based on the initial building results extracted by the image classification and verification. The best fitting circumscribed rectangles consistent with the building axis inclination were then obtained. Subsequently, the contour lines of the building polygon and their corresponding best circumscribed rectangular boundaries were divided into equal segments. After which, the Hausdorff distances between the building polygon and rectangular boundary segments were calculated. Contour preliminary regularization and optimization were accomplished herein by replacing the building segments, whose Hausdorff distances were smaller and satisfied with the replacement condition, with the corresponding best-fitting circumscribed boundary segments. Deep optimization for a complex partial contour not well optimized in the former step was also explored. The feature corner points on the complex boundary were extracted, matched, sorted, and removed to keep the best ones based on the Shi--Tomasi algorithm. Lastly, the remaining points were connected and reconstructed to optimize the complex local contour. As a result, the edge expression degree accuracy and the extraction accuracy after the contour optimization were improved. The experimental comparison and analysis results of multiple remote sensing images show that this method is not only suitable for the contour optimization of building results extracted by different classification methods but also effectively improves the edge expression accuracy of building contours. According to the change of complex building contour details, the proposed hierarchical optimization method was more accurately adaptive than the reference contour optimization method. It also achieved a better overall optimization accuracy. In other words, the building edge accuracy and regularity can be effectively improved, and the true building shape can be more accurately reflected.