关键词: 区域合并, 分水岭变换, 城市, 分层合并, 高分辨率影像

Abstract: Given that watershed transform based image segmentation methods usually produce obvious over-segmentation, and the high resolution imagery of urban areas shows some peculiar characteristics, the authors propose a hierarchical region merging method to optimize the segmentation results initially produced by watershed transformation. A multi-channel watershed transformation was adopted to generate an initial segmentation result. Through quantitative analysis of internal spectral variability of different ground objects in urban areas, the image was stratified to several layers. Region merging was separately conducted in each layer. A final segmentation result was obtained by aggregating segmentation results from these layers. The proposed method was evaluated by comparing with the existing watershed segmentation method in terms of visual inspection, quantitative measures and applications in urban land cover classification, using a QuickBird image of Beijing area. The experimental results indicate that the proposed method outperforms the existing method and it is suitable for segmentation of high resolution imagery over urban areas.

Key words: region merging, watershed transformation, urban, hierarchical merging, high resolution imagery