A Perceptual Space for Describing Human Bodies

A Perceptual Space for Describing Human Bodies
Matthew Q. Hill, Carina A. Hahn, Alice J. O’Toole
The University of Texas at Dallas
height height 3 female shapes vs. “other” male shapes vs. “other” 4 masculine vs. curvy waist height 5 waist height toned vs. average Background
Contributing descriptor terms
•  Adaptation and Norm-based Coding Studies of Body Perception
•  identity aftereffects within two bodies (Rhodes, Jeffery, Boeing, & Calder, 2013)
•  weight & gender adaptation invariant for viewpoint and pose
Female Bodies: Axes 1-­‐4 AXIS •  virtual bodies from space of 2000 laser scans (Sekunova, et al., 2013)
axis 2 tall, long, long legs skinny, round (apple), lean, peDte, small, fit axis 3 pear-­‐shaped, curvy masculine, rectangular, average, broad shoulders, muscular, long torso, short legs axis 4 curvy long torso, pear-­‐shaped, short legs axis 5 body size terms •  big, small, short, tall, heavyset, stocky, skinny, peDte global shape terms •  round (apple), rectangular, long, pear-­‐shaped, curvy fitness terms •  lean, fit, muscular, built, sturdy local feature terms •  long legs, short legs, long torso, short torso, broad shoulders averageness terms gender terms •  masculine, feminine More heavy More ‘other’ “’Female shapes vs. ‘other’” More female skinny, lean, fit short, average, short legs, small, feminine, short torso axis 2 tall, big, fit, muscular skinny, long, small, long legs axis 3 muscular, built, fit long torso, short legs, skinny, average axis 4 short torso, long legs fit, muscular, small, built, lean, skinny, short torso axis 5 average, tall, long Lower waist axis 1 Participants
•  12 undergraduate students (6 female)
•  each participant gave open ended ratings of bodies used in main experiment
•  use space to find real “opposite” bodies (in progress)
“Waist height” heavyset, round (apple), big, stocky Method
Rating scale
1.  does not apply
2.  applies somewhat
3.  applies perfectly
•  test perceptual validity of body space using adaptation paradigm
AXIS Feature Term Pilot Study
Body Feature Descriptors
Higher waist •  2 images per identity: one standing, one
walking (448 images total)
•  blurred to obscure facial identity
•  each participant rated 75 identities on 27
feature descriptors
•  total 2,025 judgments
short torso, long legs “Weight” •  rating data can be used to generate verbal descriptions when
physical measurements are known
Component 3: 10.78% Component 4: 7.43% Harms, Snow, Hurst, Pappas, & Abdi, 2005)
•  multivariate technique similar to
principal component analysis
•  used for categorical rather than
continuous variables
•  visualization of cross-tabular
•  simultaneous visualization of
observations and variables
Component 1: 38.41% More skinny •  space can be applied to generate similarity measures from verbal
descriptions of bodies
Male Bodies: Axes 1-­‐4 Correspondence Analysis
•  60 undergraduate students (30 male)
•  224 identities: 164 female, 60 male (O’Toole, •  average, proporDoned short, short legs, small, peDte, pear-­‐shaped Shorter Method
skinny, lean, proporDoned “Height” •  participants rate the applicability of body descriptor terms to a large number of bodies
•  correspondence analysis (Greenacre, 2010) separately to male and female body descriptions
•  shared perceptual body and feature descriptor space: male and female bodies
•  enables visualization of feature terms and bodies in common space
axis 1 Taller Approach
big, heavyset, round (apple), stocky Component 2: 14.59% •  silhouettes of bodies yield gender adaptation aftereffects (Palumbo, Laeng, & Tommasi, 2013)
•  rectangle width adaptation does not explain weight adaptation (Hummel, et al., 2012)
More curvy 2 •  common and gender-specific components for male and female
body spaces
Taller weight “Masculine vs. Curvy” •  map body descriptions onto perceptual and physical body spaces
weight More masculine •  (e.g., from laser scans of bodies, Freifeld & Black, 2012)
1 •  resulting spaces interpretable in the context of the feature terms
Component 4: 6.73% •  relate perceptual body spaces to physical body spaces
Male “Height” Long Term Goals
Female Shorter •  create multidimensional representation of human body shapes based on perceptual
•  map shape variation across individual bodies using body feature descriptors
•  represent both bodies and body descriptor terms in a common multidimensional space
Axis Component 2: 12.02% Goals
•  possible to reverse engineer a body similarity space from body
feature descriptors
females •  people commonly describe bodies using descriptors (e.g. skinny, curvy, heavyset,
stocky, fit, muscular, built, petite) – Can descriptions be used to reverse engineer a
representational space to describe body similarities?
males Problem
Feature selection
Exploratory Analysis
•  categories:
•  size, global shape, fitness, local feature, averageness, gender
•  descriptor terms chosen based on:
•  frequency of use within each category
Component 1: 31.88% More skinny “Weight” More heavy Component 3: 13.46% More male “Male shapes vs. More ‘other’ ‘other’” Freifeld, O., & Black, M. J. (2012). Lie bodies: a manifold representaDon of 3D human shape. In Computer Vision–ECCV 2012 (pp. 1-­‐14). Springer Berlin Heidelberg. Greenacre, M. J. (2010). Correspondence analysis. Wiley Interdisciplinary Reviews: Computa;onal Sta;s;cs, 2(5), 613-­‐619. Hummel, D., Grabhorn, R., & Mohr, H. M. (2012). Body-­‐shape adaptaDon cannot be explained by adaptaDon to narrow and wide rectangles. Percep;on, 41(11), 1315-­‐1322. O’Toole, A. J., Harms, J., Snow, S. L., Hurst, D. R., Pappas, M. R. & Abdi, H. (2005). A video database of moving faces and people. IEEE Transac;ons on PaBern Analysis and Machine Intelligence, 27(5), 812-­‐816. Palumbo, R., Laeng, B., & Tommasi, L. (2013). Gender-­‐specific aaereffects following adaptaDon to silhoueces of human bodies. Visual Cogni;on, 21(1), 1-­‐12. Rhodes, G., Jeffery, L., Boeing, A., & Calder, A. J. (2013). Visual coding of human bodies: Perceptual aaereffects reveal norm-­‐based, opponent coding of body idenDty. Journal Of Experimental Psychology: Human Percep;on And Performance, 39(2), 313-­‐317. Sekunova, A., Black, M., Parkinson, L., & Barton, J. S. (2013). Viewpoint and pose in body-­‐form adaptaDon. Percep;on, 42(2), 176-­‐186.