Relationship between KANSEI Words Describing the Human Body and Body Dimensions for Modeling Synthetic Actors

In order to design a digital human model for computer graphics based on natural language, a mathematical model which estimates the scores of KANSEI words (overweight, thin, muscular etc.) from body dimensions was developed. Evaluation observers, consisting of 49 females and 64 males, watched image photographs of 24 young women and replied semantic scores of 21 KANSEI words for each photographed subject. The 24 photographed subjects were selected from over 200 subjects based on the somatotype. Analyzing the relationship between the scores and the 35 body dimensions of the photographed subjects, it was found that the sex differences of the observers were significant. Female observers evaluated body forms in greater detail and selected the dimensions to evaluate the scores of KANSEI words more carefully. Thus, we have developed a mathematical model, which estimates the scores of KANSEI words from body dimensions for the typical female observer. Since the scores of the 21 KANSEI words were correlated with each other, an independent set of 8 KANSEI words were selected. The mathematical model was constructed for 3 KANSEI words (overweight, plump and thin) in a multivariate regression model, and a fuzzy reasoning model for the 5 other words (skinny, medium, muscular, sturdy and tight). The errors of the estimated scores of KANSEI words were validated by another 38 female observers. The errors were 5% through 20%, and it was smaller than the standard deviation of inter-observer errors.