Modelling Facial Expressions through Shape Polynomial Regression

This paper deals with the statistical analysis of landmark data observed at different temporal instants. As a motivating example, we consider the modelling of facial expressions which has great relevance in sociological studies. We work directly with offset-normal shape distributions as probability models for statistical inference in the framework of dynamic shape analysis. An Expectation Maximization (EM) algorithm for computing exact maximum likelihood (ML) estimation of the involved parameters is described and the use of a mixture of offset normal shape distributions is discussed for classification purposes.