Testing the Utility of a Data-Driven Approach for Assessing BMI from Face Images

Karin Marie Brandt Wolffhechel, Amanda C. Hahn, Hanne Østergaard Jarmer, Claire I. Fisher, Benedict C. Jones, Lisa M. DeBruine

    Research output: Contribution to journalJournal articleResearchpeer-review

    354 Downloads (Pure)

    Abstract

    Several lines of evidence suggest that facial cues of adiposity may be important for human social interaction. However, tests for quantifiable cues of body mass index (BMI) in the face have examined only a small number of facial proportions and these proportions were found to have relatively low predictive power. Here we employed a data-driven approach in which statistical models were built using principal components (PCs) derived from objectively defined shape and color characteristics in face images. The predictive power of these models was then compared with models based on previously studied facial proportions (perimeter-to-area ratio, width-to-height ratio, and cheek-to-jaw width). Models based on 2D shape-only PCs, color-only PCs, and 2D shape and color PCs combined each performed significantly and substantially better than models based on one or more of the previously studied facial proportions. A non-linear PC model considering both 2D shape and color PCs was the best predictor of BMI. These results highlight the utility of a "bottom-up", data-driven approach for assessing BMI from face images.
    Original languageEnglish
    Article numbere0140347
    JournalP L o S One
    Volume10
    Issue number10
    Number of pages10
    ISSN1932-6203
    DOIs
    Publication statusPublished - 2015

    Bibliographical note

    This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

    Fingerprint

    Dive into the research topics of 'Testing the Utility of a Data-Driven Approach for Assessing BMI from Face Images'. Together they form a unique fingerprint.

    Cite this