PDF version of this document
A very common use of AAM's is for medical image analysis and face recognition. Active appearance models possess traits that make them robust and effective in the biological domain, whereas industrial inspection, for example, presents some inherently different problems. These problems are often solved more quickly by other approaches that are based on lower-level knowledge about the image contents. Since a broad range of tasks are performed in industrial inspection, however, it is also valid to assume that the suitability of top-down approach is irrespective of the problem.
In order to visualise biological shapes and full appearances, a model which handles anatomical variability and change needs to be used. It must account for natural or pathological changes such as the change in form of organs. Greater variability can be encountered when aligned images are obtained from different subjects in a population (inter-subject), the same subject at different time instances (or different sites) or when having to account for movement such as the that which occurs due to respiration, the cardiac cycle, etc. A separate case to consider is multi-modal imaging which will not be explained in any detail although it is a fast-developing area.
For a good overview on many of the different image analysis techniques, Sonka [] is a valuable source. For a good review of model-based image analysis, papers from Cootes et al. are an even better source although the ideas are partially based on concepts such as snakes, bending-energy and active contour models that led to the development of active shape models. Sonka discusses all these in depth.