
While 3D body models have been vastly studied in the last decade, acquiring accurate models from the sparse information about the subject and few computational resources is still a main open challenge. In this paper, we propose a methodology for finding the most relevant anthropometric measurements and facial shape features for the prediction of the shape of an arbitrary segmented body part. For the evaluation, we selected 12 features that are easy to obtain or measure including age, gender, weight and height; and augmented them with shape parameters extracted from 3D facial scans. For each subset of features, with and without facial parameters, we predicted the shape of 5 segmented body parts using linear and non-linear regression models. The results show that the modeling approach is effective and giving sub cm reconstruction accuracy. Moreover, adding face shape features always significantly improves the prediction.