Building Extraction and Graph-cut based Optimization Via Frame Field Representation Learning

Building extraction and graph-cut based optimization via frame field representation learning is proposed to address the issue that conventional methods solely relying on gray convolution operation cannot obtain an effectively structured feature description of the building. First, following multi-threshold segmentation by HSV, the texture information in the high-resolution remote sensing image is fused with the elevation information in DSM. In order to reinforce the constraints of building structure information, pixel vector information is simultaneously added to the segmentation process. The frame field and preliminary segmentation results are then obtained. After region segmentation yields the basic nodes of the graph-cut method, the fusion of different features is used to perform binary segmentation based on graph-cut. Finally, iteration is used to optimize the frame field and segmentation results. The results of the experiment demonstrate that this algorithm’s segmentation enhances overall accuracy and precision and that the extracted building structure is more comprehensive and has better detail expression than other models.