A method for segmenting at least a pair of regions of an image, such method comprising: obtaining data of the image; computing watersheds of the image from intensity gradients of such image data; extracting a watershed region adjacency graph from the computed watersheds, such graph comprising a plurality of nodes corresponding to the watersheds and node interconnecting edges; assigning weights to the interconnecting edges; identifying each of the pair of regions in the image; identifying the nodes corresponding to the pair of identified regions in the adjacency graph; applying constrained graph-partitioning in the adjacency graph using the edge-weights to label unmarked nodes corresponding to each one of the pair of regions; and extrapolating the obtained label nodes on the graph to the image to segment each one of the pair of regions of the image.