Reparameterising 3D Statistical Shape Models

摘要

3D statistical shape models are widely used in modelling 3D shapes such as human faces and bodies. The limitation of such model is that, once built, the model can only represent 3D shape instances of a fixed mesh topology. While some applications may require a shape model of a different mesh topology, the model building pipeline has to be repeated with the new template, which could be time and computational resource consuming. In other cases only the statistical model is available and access to the original data is not possible. In this paper, we present a method to reparameterise a given 3D statistical shape model to any topology without using any training data. We also show that the reparameterised model achieves comparable performance as the original model.