Active Appearance Models applied to Human Faces.

The tutorial will give an overview of the active appearance model algorithm (AAM). The AAM is
a generative model of a class of objects learnt from manually labelled training data. The
method combines shape and texture variation to produce a generic model which can be fitted
to new examples of the object which resemble the training data. The method is generic and
has been very popular in the Computer Vision field over the past 8 to 9 years, since the method
was first described.

Applications to human faces and medical image data will be presented. Various
extensions of the AAM will be described for example varying the texture representation and
update method used when fitting the model. Results of face
recognition experiments using the AAM framework will be discussed, highlighting the importance
of accurate model fitting.

The second part of the tutorial will describe the practical application of appearance models to
human faces in a fully automatic system. I will present a live demonstration which consist of
four elements, namely face detection, feature localisation, local model fitting and face verification.
I will also present the Constrained Local Model algorithm (CLM). A method which uses a training method
similar to the AAM, but uses a novel fitting method. The tutorial will conclude with a brief summary
of recent approaches to automatically registering images, to avoid the labour intensive
manual landmarking of training data.