The precise location and identification of buildings are of importance to many geospatial applications. High-resolution satellite images with multispectral channels contain abundant spectral and structural information about ground objects, making these images more suitable for automatic building detection. However, the automatic detection of buildings is still very difficult owing to many obstacles, such as different imaging conditions, complex background, and various types of buildings. Therefore, this paper proposes a novel hierarchical building extraction method based on object-oriented and morphological models for automatic building detection from high-resolution satellite images captured in complex scenes.
The proposed method first extracts build-up areas from high-resolution satellite images and then detects buildings from the extracted build-up areas. In the procedure of build-up areas extraction, the multi-scale and multi-directional Gabor wavelet transform is first applied to high-resolution satellite images. Then, a scale-invariant feature point detection algorithm that considers the multi-scale and multi-directional texture properties of build-up areas is proposed for the detection of building feature points. Subsequently, watershed segmentation algorithm with threshold mark is utilized to obtain homogeneous regions, and a spatial voting matrix is computed based on these homogeneous regions and the detected feature points to obtain confidence map. Finally, build-up areas are extracted by segmenting the confidence map using adaptive thresholding algorithm. In the procedure of building extraction, the morphological building index (MBI) is first applied to the extracted build-up areas, and then the initial building results are obtained by performing threshold segmentation on the MBI feature image. Finally, shape attributes such as length-width ratio are used to further refine the initial building extraction results.
The performance of the proposed method is evaluated using three high-resolution satellite images captured in complex environments. Evaluation results show that the proposed method can efficiently and accurately detect buildings in complex scenes with an overall accuracy and Kappa coefficient greater than 90% and 0.8, respectively. The proposed method also improves the omission and commission errors by 10.03% and 6.86% on average, respectively, as compared with the performance of the PanTex algorithm.
A novel hierarchical building extraction method based on object-oriented and morphological models is proposed in this study. The experimental results highlight the advantages of the hierarchical extraction strategy and demonstrate that the proposed method outperforms the PanTex algorithm. However, good performance of the proposed method relies heavily on the detection of built-up areas, and further improvements should be performed in the future.