Automated Building Segmentation in Areal Images Using Boundary Edge Detection

Automated building extraction from satellite images is widely used for urban planning, disaster management, and many other applications, which is a challenging research problem due to its variability. Edge information obtained from satellite images is greatly affected by varying image contrast due to non-uniform illumination and resolution, resulting in the absence of edges, the presence of false edges, longer unbroken edges, and fragmented edges. Furthermore, edges play a crucial role in identifying urban objects, roads, and buildings. To address these challenges, this research paper introduces an algorithm to automatically extract urban buildings from high-resolution satellite imagery using edges as key features. The proposed method, Two-Pass Building Detection (TPBD), works in two steps: (i) Bounding Box Identification is used to locate candidate buildings using local feature points obtained by the Harris corner detector. (ii) An edge-based segmentation method is then applied to the bounding box to classify edges. Then, based on the shape features, horizontal and vertical edge pairs of buildings are identified. By iterating through a wide parameter space using various localization filters, the algorithm extracts accurate boundary edges. The effectiveness of the proposed algorithm is evaluated by performing experiments on the SztaKi–Inria dataset. The results demonstrate that the proposed algorithm achieves a remarkable accuracy rate of 93% on this benchmark dataset.