Abstract
Human body shape variations can be measured physically and described verbally (e.g., curvy, stocky), though little is known about the relationship between body shapes and descriptions. We examined this relationship by linking two similarity spaces: one created from body descriptions (descriptor space) and the other from full-body laser scans (shape space). The descriptor space was generated from a multivariate correspondence analysis (CA) applied to participants’ ratings of 164 female bodies using 27 body-descriptive terms. The shape space was made using a database of approximately 2000 full-body laser scans (cf. Anguelov et al. 2005). In interpreting individual axes, it appeared that similar information was captured in the first five components of both spaces, although axis ranking differed. To compare the spaces, the factor scores for the 164 bodies from the descriptor space (real bodies) were scaled and projected into the shape space, manually changing axis ranks according to interpretation. Synthetic bodies were created at these locations in the shape space. The resemblance between the synthetic and real bodies was tested as follows. Participants (n=60) rated the synthetic bodies using the original descriptor terms, in order to project them into the descriptor space. Next we measured the Euclidean distance between the real and synthetic body projections. The results indicated a resemblance between the real and synthetic bodies. First, matched pairs of real and synthetic bodies were closer, on average, than non-matched pairs, with a bootstrap test showing no overlap between the matched and non-matched mean distance distributions. Second, the coordinates of matched pairs were highly correlated (Axes 1-5; 0.94, 0.79, 0.71, 0.26, 0.32, all p’s < 0.0008). We conclude that the verbal descriptions encapsulate critical dimensions of body shape. The combined space makes it possible to generate approximated body shapes from verbal descriptions and to generate approximated descriptions for arbitrary body shapes.
Meeting abstract presented at VSS 2015