5D Library › Xuankong Flying-Star Permutations, Twenty-Four-Mountain Compass Discretization and Craniofacial Ratio Metrology · 中文

Chapter 4. Thirds, Five Eyes, the Golden Ratio and Selfie Perspective

Base rates of ideal proportions

Physiognomy often describes an ideal face by its “three thirds” and “five eyes”: hairline to brow, brow to nose base and nose base to chin of equal length, and a face as wide as five eyes. Popular accounts also treat the golden ratio φ\varphi as the measure of beauty and nobility. Anthropometry measures such dimensions between standard landmarks and finds considerable variation between people (Farkas 1994), and the ubiquity of the golden ratio in nature and the body is often exaggerated (Livio 2002).

How many people meet an ideal proportion? The program takes an illustrative population: upper, middle and lower thirds with means 6.0, 6.0 and 6.5 cm and coefficient of variation cc; face width over eye width with mean 4.6 and sd 0.3; face height over width with mean 1.35 and sd 0.08. “Equal thirds” means the longest third is at most 1.05 times the shortest, “five eyes” a ratio within 5 ± 0.1, and a “golden face” a height-to-width ratio within 2% of φ\varphi.

Calculation

# Chapter 4. p07: How rare are "ideal" thirds, fifths and golden faces?
# Illustrative population of faces (lengths in cm, independent normal):
#   upper third (hairline to brow) mean 6.0, middle third (brow to
#   nose base) mean 6.0, lower third (nose base to chin) mean 6.5,
#   each with coefficient of variation c;
#   face width / eye width ("five eyes") mean 4.6, sd 0.3;
#   face height / face width mean 1.35, sd 0.08.
# "Equal thirds" means the longest third is at most 5 % longer than
# the shortest; "five eyes" means the ratio is within 5 +/- 0.1;
# "golden" means height / width within 2 % of phi = 1.618.
# 200,000 simulated faces, fixed seed.
import math
import random

PHI = (1 + math.sqrt(5)) / 2

if __name__ == "__main__":
    rng = random.Random(1618)
    n = 200000
    print("   phi = {:.6f}".format(PHI))
    print("   CV of thirds   equal thirds   five eyes   golden   all three")
    for c in (0.06, 0.08, 0.10):
        eq = fe = go = al = 0
        for _ in range(n):
            t = [rng.gauss(m, m * c) for m in (6.0, 6.0, 6.5)]
            a = max(t) / min(t) <= 1.05
            b = abs(rng.gauss(4.6, 0.3) - 5) <= 0.1
            g = abs(rng.gauss(1.35, 0.08) / PHI - 1) <= 0.02
            eq += a
            fe += b
            go += g
            al += a and b and g
        print("   {:12.2f} {:14.4f} {:11.4f} {:8.4f} {:11.5f}".format(
            c, eq / n, fe / n, go / n, al / n))
    z = (PHI * 0.98 - 1.35) / 0.08
    print("   golden band starts {:.2f} sd above the mean height / width"
          .format(z))
    print("\nReading: under these illustrative spreads, equal thirds")
    print("are uncommon, the five-eye proportion is met by about one")
    print("face in ten, and a golden face is almost absent. An ideal")
    print("that almost nobody meets labels nearly everyone a deviation,")
    print("which leaves room to read meaning into any face.")
#    phi = 1.618034
#    CV of thirds   equal thirds   five eyes   golden   all three
#            0.06         0.0988      0.1122   0.0014     0.00002
#            0.08         0.0711      0.1104   0.0014     0.00001
#            0.10         0.0508      0.1114   0.0015     0.00001
#    golden band starts 2.95 sd above the mean height / width
#
# Reading: under these illustrative spreads, equal thirds
# are uncommon, the five-eye proportion is met by about one
# face in ten, and a golden face is almost absent. An ideal
# that almost nobody meets labels nearly everyone a deviation,
# which leaves room to read meaning into any face.

With a coefficient of variation of 0.06, only 0.0988 of faces have equal thirds, and 0.0508 at 0.10; about one in nine meets the five-eye proportion; golden faces number only about 0.0014, because the golden band begins 2.95 standard deviations above the population mean. One to two faces in a hundred thousand meet all three. An ideal almost nobody meets labels almost everyone a “deviation”, which leaves room to read meaning into any face.

A selfie enlarges the nose

A pinhole camera images an object at distance dd at a scale proportional to 1/d1/d. The nose tip lies about 2.5 cm in front of the reference plane through the outer eye corners and the cheekbone points about 5 cm behind it (illustrative depths), so a photo inflates the ratio of nose width to cheekbone width by (d+5)/(d−2.5)(d + 5)/(d – 2.5). A published study computed the same effect with a model: a selfie taken at about 12 inches makes the nasal base look about thirty percent wider than at about 5 feet (Ward et al. 2018).

Calculation

# Chapter 4. p08: A selfie enlarges the nose: perspective magnification.
# A pinhole camera at distance d (cm) from a reference plane through
# the outer corners of the eyes images a point at depth z (cm, toward
# the camera positive) with scale f / (d - z). The nose tip lies about
# 2.5 cm in front of that plane, the cheekbones (zygion) about 5 cm
# behind it (illustrative depths). The apparent ratio of nose width to
# cheekbone width is inflated by the factor (d + 5) / (d - 2.5).
# Ratios read from the photo are compared with the true ratio, and the
# distance needed to keep the distortion under 2 % is solved exactly:
#   (d + 5) / (d - 2.5) <= 1.02  <=>  d >= (5 + 1.02 x 2.5) / 0.02.

def factor(d):
    return (d + 5.0) / (d - 2.5)

if __name__ == "__main__":
    print("   camera distance (cm)   nose / cheek ratio inflated by")
    for d in (25, 30, 45, 60, 100, 150, 200, 300):
        print("   {:20d} {:22.3f}".format(d, factor(d)))
    print("\n   distance for distortion under 2 %: {:.1f} cm".format(
        (5 + 1.02 * 2.5) / 0.02))
    true_ratio = 0.27
    print("   a true nose / cheek ratio of {:.2f} reads {:.3f} at 30 cm"
          .format(true_ratio, true_ratio * factor(30)))
    print("   and {:.3f} at 200 cm".format(true_ratio * factor(200)))
    print("\nReading: an arm's-length selfie inflates the nose against")
    print("the cheekbones by more than a quarter. Any proportion read")
    print("from a phone photo mixes the face with the camera distance;")
    print("only photos taken from a few metres compare faces fairly.")
#    camera distance (cm)   nose / cheek ratio inflated by
#                      25                  1.333
#                      30                  1.273
#                      45                  1.176
#                      60                  1.130
#                     100                  1.077
#                     150                  1.051
#                     200                  1.038
#                     300                  1.025
#
#    distance for distortion under 2 %: 377.5 cm
#    a true nose / cheek ratio of 0.27 reads 0.344 at 30 cm
#    and 0.280 at 200 cm
#
# Reading: an arm's-length selfie inflates the nose against
# the cheekbones by more than a quarter. Any proportion read
# from a phone photo mixes the face with the camera distance;
# only photos taken from a few metres compare faces fairly.

At 30 cm the inflation is 1.273, at 25 cm 1.333, and even at 200 cm 1.038. Keeping the distortion under 2% needs the camera beyond 377.5 cm. A face with a true ratio of 0.27 reads 0.344 in a 30 cm selfie and 0.280 at 200 cm.

Perspective inflation of the nose-to-cheekbone ratio against camera distance. Dotted lines: 2% and 377.5 cm.

Hypothesis · not established

Testable consequence. If a face reading really comes from facial structure, two photos of the same person taken at 30 cm and at 4 m should yield the same reading. What would show this wrong: if readings shift systematically with camera distance (for instance, “large nose” more often in close-ups), the reading comes at least partly from the lens and not the face.

Summary

Equal thirds, five eyes and golden faces are ideals few people meet, and golden faces hardly exist; the rarer the ideal, the easier it is to read anyone as “distinctive”. A phone selfie inflates the nose against the cheekbones by more than a quarter, so camera distance must be controlled whenever proportions are read from photos.

References

Farkas, Leslie G., ed. 1994. Anthropometry of the Head and Face. 2nd ed. Raven Press.
Livio, Mario. 2002. The Golden Ratio: The Story of Phi, the World’s Most Astonishing Number. Broadway Books.
Ward, Brittany, Max Ward, Ohad Fried, and Boris Paskhover. 2018. “Nasal Distortion in Short-Distance Photographs: The Selfie Effect.” JAMA Facial Plastic Surgery 20 (4): 333–35.

Fan, Y. (2026). Xuankong Flying-Star Permutations, Twenty-Four-Mountain Compass Discretization and Craniofacial Ratio Metrology: Lo-Shu Orbit Cycles, Magnetic Declination Flip Rates and Landmark Measurement Error Propagation [玄空飞星置换代数、二十四山罗盘离散化与骨相比例计量] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.24090215 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).