The marginalisation property

Gaussian Processes provide us with a mathematically elegant way of modelling shape deformations. As shape modelling is an application-oriented task, we are not primarily interested in mathematical elegance, but rather in obtaining practical algorithms.

Thanks to the marginalisation property, Gaussian Processes also satisfy this criterion. We will discuss how the marginalisation property allows us to obtain a discrete representation of the shape variation in terms of a multivariate normal distribution and how we can use this to obtain interesting information about the shapes that are represented by the model.