A multi-label interactive image segmentation method based on region merging

Interactive image segmentation is becoming more and more popular. In this paper we describe a multi-label interactive image segmentation method which incorporates both seeded-region extension and region merging. First, Mean shift method is provided for the initial segmented regions, and then the users draw lines to mark the position and region of the foreground and background, following that, foreground regions expand based on the similarity with non-marker regions and foreground regions according to the ordered cluster, finally, based on maximal-similarity region merging strategy, we merge the initial regions to get the final segmentation result. The results of experiments show that with this method users could get a good segmentation result for most multi-cell images, also can revise the results and re-segment to get an ideal result.