File:Njp495117f2 hr.jpg

English: Identifying cell centers. (A) Illustration in Euclidean space: in order to separate the black dots into clusters we partition the plane into 12 × 25 squares and measure the local density in each square as the number of dots included. This is indicated by the color on a scale from green (0 density) to red (largest is 6 dots/square). A square becomes a generator point (white star) if it has the largest density in its neighborhood (gray circle) with radii r = 4 (top) and r = 7 (bottom) (one unit is the edge of one square). The Voronoi partitioning is shown with blue lines. (B) Generalizing the algorithm to graphs, each node will have a local density (indicated by the size of the nodes and also the height of the small hills above them). The length of the edges is proportional to the inverse of the edge clustering coefficient (shown by the width of the edges). Generator nodes (the value of their local density is shown) have the highest density in their neighborhood with radius r1<r2<r3. When increasing r, clusters are sequentially merged together.