Building Vector Contour Extraction from Remote Sensing Images Based on Multi-Level Contour Refinement and Morphological Perception

Accurate extraction of building vector contours from high-resolution remote sensing images is a fundamental task for urban mapping and geographic information systems. However, existing approaches often suffer from blurred boundaries and geometric distortions when dealing with buildings of complex shapes, limiting the accuracy and usability of the extracted building footprints. To address these challenges, this paper proposes a multi-level building contour refinement framework based on morphological perception. The proposed framework integrates a three-stage contour optimization strategy, including principal direction extraction, morphology-based contour reconstruction, and geometry-aware regularization, to progressively refine complex building contours under geometric constraints. In addition, a multi-dimensional contour complexity model and an adaptive threshold optimization network are introduced to dynamically adjust refinement parameters according to contour complexity. Experimental results on the WHU-Mix dataset demonstrate that the proposed method outperforms state-of-the-art approaches, achieving 87.52%, 77.43%, and 87.35% in boundary F1, vertex F1, and mIoU, respectively. These results indicate that the proposed framework provides an effective and robust solution for high-precision building vector contour extraction in complex remote sensing scenarios.