Skip to Main Content
In this paper, a two-step segmentation algorithm is proposed based on watershed transform to segment multi-spectral satellite images. The first step is to use watershed segmentation to gain the initial over-segmented regions and the next one is region merging using a strategy of minimizing the overall heterogeneity increased within segments at each merging step. Textural, color and shape information of segments is used in the merging process. The study was conducted to explore an efficient approach to segment remote sensing images especially for high resolution multi-spectral satellite imagery. Experimental results show that the proposed method can produce quite good segmentation results and is very promising in segmentation of remotely sensing imagery in the future.