Quantifying Shape Variation with Procrustes Alignment

49 concepts46 papers2 proposals

Imagine trying to compare the physical structure of two skulls, two leaves, or two human faces. If one image is zoomed in, tilted sideways, or shifted to the left, a direct pixel-by-pixel or coordinate-by-coordinate comparison will fail.

Historically, researchers relied on linear morphometrics—measuring simple distances like height, width, or inter-ocular ratios. However, as Diana Weiting Tan et al. demonstrate in A broad autism phenotype expressed in facial morphology, linear ratios discard the rich, non-linear geometric relationships of a structure. To truly analyze morphology, we need a way to isolate pure shape—the geometric information that remains once position, scale, and orientation are completely stripped away.

Landmark-based geometric morphometrics solves this problem by mapping structures to a shared coordinate space using a process called Procrustes alignment.

The Landmark Framework

The foundation of this approach relies on identifying a standardized set of anatomical points, known as landmarks. As Qianhong Gu et al. show in Phylogeography of Bellamya (Mollusca: Gastropoda: Viviparidae) snails on different continents: contrasting patterns of diversification in China and East Africa, mapping spatial coordinates of specific anatomical points allows for the mathematical isolation of shape changes.

For these landmarks to be useful, they must be homologous—meaning they represent the exact same biological or structural feature across every specimen in a study. For example, in evolutionary biology, Thomas Cucchi et al. use 2D coordinates anchored to specific dental features in Social Complexification and Pig (Sus scrofa) Husbandry in Ancient China: A Combined Geometric Morphometric and Isotopic Approach to track subtle morphological shifts caused by domestication.

[Raw Coordinates] ──> [Translation] ──> [Scaling] ──> [Rotation] ──> [Pure Shape Coordinates]

The Core Engine: Generalized Procrustes Analysis

Once landmarks are recorded, they still contain confounding extrinsic variations: where the object was positioned relative to the camera, how large the organism grew, and how it was rotated.

To eliminate these variables, researchers use Generalized Procrustes Analysis (GPA). As detailed by Marine Jeanjean et al. in Sorting the flock: Quantitative identification of sheep and goat from isolated third lower molars and mandibles through geometric morphometrics, GPA standardizes shape data through a three-step geometric normalization process:

  1. Translation (Centering): The center of mass (centroid) of each landmark configuration is calculated. Every configuration is then shifted so that its centroid sits precisely at the origin $(0,0,0)$.
  2. Scaling (Size Normalization): To ensure that absolute size does not skew the analysis, configurations are scaled to a common unit. Washington Mio et al. describe this in Modeling Brain Anatomy with 3D Arrangements of Curves as scaling configurations to a unit sphere, ensuring that the calculated distance between two structures is invariant to their absolute size.
  3. Rotation (Orientation Alignment): The configurations are iteratively rotated around the origin until the sum of squared distances between corresponding landmarks across all specimens is minimized.

By removing these global variations, the model can focus exclusively on internal deformations. As Eric P. Xing et al. note in Learning GMRF Structures for Spatial Priors, this canonical shape normalization allows statistical models to focus entirely on local spatial relationships between object components.

Lauren Finka demonstrates the power of this isolation in Geometric morphometrics for the study of facial expressions in non-human animals, using the domestic cat as an exemplar, showing that shape-invariant coordinate normalization anchors all configurations to a shared reference frame, separating dynamic expressions from the subject's distance or angle relative to the sensor.

Handling Continuous Contours

Discrete landmarks work well for sharp intersections or distinct points, but biological structures often feature smooth, continuous curves lacking obvious point-like features. To handle this, the framework expands in two ways:

  • Sliding Semi-Landmarks: As utilized by Marine Jeanjean et al., points are placed along a curve and allowed to "slide" during the alignment phase. This minimizes artificial shape differences that would otherwise be caused by placing points arbitrarily along a smooth edge.
  • Elliptic Fourier Analysis: When dealing with entirely continuous boundaries, such as tooth outlines, Greger Larson et al. demonstrate in Phylogeny and ancient DNA of Sus provides insights into neolithic expansion in Island Southeast Asia and Oceania how decomposing a shape into a series of harmonic components allows for objective mathematical comparison without discrete landmarks.

From Geometry to Statistical Inference

Once Procrustes alignment outputs "pure shape" coordinates, these coordinates can be treated as standard features for downstream statistical modeling and machine learning.

For instance, Giovanni Maria Farinella et al. leverage this in Assessment of Cow's Body Condition Score Through Statistical Shape Analysis and Regression Machines by applying Kernel Principal Component Analysis (KPCA) to aligned landmarks to predict body condition scores from dorsal contours. Similarly, Stanislaw Kazmierczak et al. show in Prediction of the Facial Growth Direction is Challenging that geometric landmark normalization improves the consistency of feature representations, allowing machine learning models to predict how a child's face will structurally mature.

Ultimately, by converting raw spatial positions into normalized geometric coordinates, Procrustes alignment provides a mathematically rigorous foundation for analyzing physical form across biology, medicine, and computer vision.

Key Foundational Work

JPA volume 49 issue S8 Cover and Front matter (1975) — contributed 1 concept elements to this direction.
Modeling Brain Anatomy with 3D Arrangements of Curves (2007) — contributed 1 concept elements to this direction.
Biometric Evidence that Sexual Selection Has Shaped the Hominin Face (2007) — contributed 1 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_geometric_morphometrics_dd00efec,
  title = {geometric morphometrics},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/geometric-morphometrics/rdr_dd00efec352a4b47b21dade60bfaa8c7},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-19},
}

This exact version (version 1, frozen text):

@misc{lacuna_geometric_morphometrics_dd00efec_v1,
  title = {geometric morphometrics},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/geometric-morphometrics/rdr_dd00efec352a4b47b21dade60bfaa8c7/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-19},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake