A benchmark for geometric facial beauty study

Fangmei Chen, Dapeng Zhang

Research output: Chapter in book / Conference proceedingConference article published in proceeding or bookAcademic researchpeer-review

26 Citations (Scopus)

Abstract

This paper presents statistical analyses for facial beauty study. A large-scale database was built, containing 23412 frontal face images, 875 of them are marked as beautiful. We focus on the geometric feature defined by a set of landmarks on faces. A normalization approach is proposed to filter out the non-shape variations - translation, rotation, and scale. The normalized features are then mapped to its tangent space, in which we conduct statistical analyses: Hotelling's T2test is applied for testing whether female and male mean faces have significant difference; Principal Component Analysis (PCA) is applied to summarize the main modes of shape variation and do dimension reduction; A criterion based on the Kullback-Leibler (KL) divergence is proposed to evaluate different hypotheses and models. The KL divergence measures the distribution difference between the beautiful group and the whole population. The results show that male and female faces come from different Gaussian distributions, but the two distributions overlap each other severely. By measuring the KL divergence, it shows that multivariate Gaussian model embodies much more beauty related information than the averageness hypothesis and the symmetry hypothesis. We hope the large-scale database and the proposed evaluation methods can serve as a benchmark for further studies.
Original languageEnglish
Title of host publicationMedical Biometrics - Second International Conference, ICMB 2010, Proceedings
Pages21-32
Number of pages12
DOIs
Publication statusPublished - 21 Jul 2010
Event2nd International Conference on Medical Biometrics, ICMB 2010 - Hong Kong, Hong Kong
Duration: 28 Jun 201030 Jun 2010

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume6165 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference2nd International Conference on Medical Biometrics, ICMB 2010
Country/TerritoryHong Kong
CityHong Kong
Period28/06/1030/06/10

ASJC Scopus subject areas

  • Theoretical Computer Science
  • Computer Science(all)

Cite this