As mentioned earlier, AAMs require a shape model, and this role is played by Active Shape Models (ASMs). In the upcoming sections, we will create an ASM that is a statistical model of shape variation. The shape model is generated through the combination of shape variations. A training set of labeled images is required, as described in the article Active Shape Models--Their Training and Application, by Timothy Cootes. In order to build a face-shape model, several images marked with points on key positions of a face are required to outline the main features. The following screenshot shows such an example:
There are 76 landmarks on a face, which are taken from the MUCT dataset. These landmarks are usually marked up by hand, and they outline several face features such as mouth contour, nose, eyes, eyebrows, and face shape, since they are easier to track.