Voronoi Diagram Modeling of Natural Patterns

What this research found

An illustrated explainer that takes one geometric rule — every point belongs to the seed nearest to it — and tests how well it actually describes patterns in nature. The same Voronoi pipeline was run over photographs of a Chinese money plant, a reticulated giraffe, and a cracked mud flat, plus a control set of 80 random points, then all four were compared on three scale-free statistics. The plant emerged as the closest natural match to an optimally even tessellation, while the giraffe and mud scores turned out to be dominated by a measurement artefact that the write-up deliberately leaves on display.

• The money plant is the cleanest natural example in the set. Its leaf centres give a Clark-Evans regularity index of 1.42, above the random control's 1.12, and a normalised seed-to-centroid offset of only 0.20 — close to a centroidal Voronoi tessellation, the arrangement that covers a region most evenly.
• The pipeline recovered the plant's packing without being told where the leaves were. Twenty-two leaf-disc centres found by a Hough circle transform on a green foliage mask produced cell edges that run along the gaps between neighbouring leaves.
• The giraffe (Clark-Evans 0.32) and the mud flat (0.49) score as clustered, but that is a support-region artefact rather than biology. Their seeds occupy only part of the photograph, so cells near the silhouette absorb empty background, pushing cell-area variation to 2.65 and 1.52 against 0.52 for the plant.
• Random points alone look convincingly organic. The 80-point uniform control matches the plant's coefficient of variation in cell area exactly at 0.52, the textbook value for a two-dimensional Poisson-Voronoi tessellation, which is why visual comparison on its own settles nothing.
• None of the natural cases is exactly Voronoi. Giraffe patches emerge from reaction-diffusion during development, and mud cracks meet mostly at T-junctions of 90, 90 and 180 degrees rather than the 120-degree triple junctions of a generic tessellation, because cracks form in sequence rather than all at once.

How it was done

Three public-domain and CC-BY-SA photographs from Wikimedia Commons — a top-down Pilea peperomioides plant, a reticulated giraffe, and a dried elephant mudwallow in Samburu National Reserve, Kenya — plus a synthetic random control were resized to at most 2000 pixels on the long side and run through one pipeline. Seed extraction was tuned per subject: colour masking with a Hough circle transform for the leaf discs, a brown-hue mask with watershed segmentation and circularity filters for the coat patches, and local thresholding of the crack network for the mud islands. Voronoi cells were computed in SciPy, infinite ridges were clipped to the working bounding box, and edge cells were made finite by mirroring seeds across each boundary before clipping. Each subject was then scored on cell-area variation, the Clark-Evans nearest-neighbour index, and seed-to-centroid offset, and the results written up as a 15-page tour pitched at a reader with one year of algebra.

Data sources

• Wikimedia Commons — top-down photograph of Pilea peperomioides foliage, CC-BY-SA 4.0 (22 leaf seeds)
• Wikimedia Commons — reticulated giraffe photograph, CC-BY-SA 4.0 (30 coat-patch seeds)
• Wikimedia Commons — dried elephant mudwallow, Samburu National Reserve, Kenya, CC-BY-SA 4.0 (21 mud-island seeds)
• Okabe, Boots, Sugihara & Chiu, Spatial Tessellations, 2nd edition (2000)
• Du, Faber & Gunzburger, SIAM Review 41:637 (1999) — centroidal Voronoi tessellations
• Goehring, Philosophical Transactions of the Royal Society A 371:20120353 (2013) — polygonal fracture patterns
• Andrews, Baccelli & Ganti, IEEE Transactions on Communications 59:3122 (2011) — Voronoi models of cellular coverage