next up previous
Next: Relationship to Original Proposal Up: Research Activity Previous: Active Appearance Models (AAMs)

Image Annotation

Perhaps the most significant difficulty with applying the models to new problems is the requirement for a set of labelled training examples. The labelling is time consuming and tedious. One of my main priorities, therefore, has been to develop algorithms and tools to speed up and perhaps eventually fully automate this process. This was not in the original project plan, but is a very important issue. Most of the research in the proposal require sets of labelled images - tools to generate such data rapidly will be of great help.

I have been writing such a tool, which allows interactive editting of label points, search using ASMs and AAMs and methods of adding extra variation to shape models to deal with small training sets. As well as reducing the time required to mark up new images, it is a useful test-bed for the search algorithms.

Tim Cootes