Segmentation of very high resolution imagery based on multichannel watershed transformation and region merging

This paper proposed a hierarchical region merging method that based on the spectral similarity, to optimize the initial segmentation results from watershed transformation, addressing the problem that the existing watershed transformation algorithms are very sensitive to noise and prone to produce over-segmentation. The proposed method was evaluated by comparing with the most widely used watershed segmentation method through visual inspection, quantitative measures and applications in urban land cover classification, using a Quickbird image data of Beijing area and achieved a comparable performance.