Human body prediction of size and shape: a hormonal framework

Abstract

To achieve high prediction accuracy with minimal inputs from online retail respondents, a method was developed and tested to predict the size and shape of the human body in 3D using a hormonal framework. The prediction method is based on geometric morphometrics, image analysis, and kernel partial least squares regression. The inputs required are answers to three closed-ended questions and a passport photo. Prediction accuracy was tested with the 3D body scan dataset of the Civilian American and European Surface Anthropometry Resource project. Results from the test dataset showed that approximately 82% of the error expectations of landmarks followed a log-normal distribution with an expectation of 8.816 mm and standard deviation of 1.180 mm. The remaining 18% of the error expectations of landmarks followed a log-normal distribution with an expectation of 18.454 mm and standard deviation of 8.844 mm, which may herald future research. Benchmarked with another method, the proposed method features much less input. In addition to high accuracy, the method in this paper allows for visualisation of results as real-size meshes in millimeters.