Active Cues Collection and Integration for Building Extraction With High-Resolution Color Remote Sensing Imagery

Building extraction from high-resolution color remote sensing imagery (HRCRSI) is important in city planning, building reconstruction, and other applications. However, the performance of the state-of-the-art methods is often passively dependent on the accuracy and reliability of the initial edges/regions acquired by general edge/region extraction methods. But the performance of these general methods is always sensitive to unavoidable noise and interferences, especially for the HRCRSI imagery. Furthermore, structural information of the target (e.g., buildings herein) is not fully utilized in these general methods, which is undoubtedly a useful clue to reducing the effects of noise and interference. Therefore, undesired results are inevitable for building extraction methods conducted with a passive or semiactive manner. In this paper, in order to alleviate this problem to a certain extent, we carried out the building extraction task in a completely active manner: 1) under the guidance of the visual perception theory, cues of building edges and regions are actively collected by considering building priors related to main direction and color; and 2) based on knowledge about building shape widely accepted in the literature, cues of the obtained building edges and regions are actively integrated for final building extraction. Experimental results on three benchmark datasets, including aerial and high-resolution optical satellite images, illustrate that the proposed active method can achieve the expected building extraction results.