An interesting and integral part of appearance model are their construction (or formation) step. The first step is concerned with the establishment of a model that not only describes a mean form of some object in an image (if not the image as a whole), but also the legal variation that can be applied to that mean in order to create new legal object instances. A model formulates the form which vectors can take and these vectors can easily be translated into a visual description. More desirable models will not be excessively data-permissive. They should allow recognition and acceptance of only reasonable variations of the object under investigation. There is a convenient mathematical way of expressing this variation and that is to assign a parameter to each mode of variation2.11. When change in these parameters occurs and the mean is deformed accordingly, there will be a direct effect on the appearance of the result. Rather usefully, each legal instance can always be uniquely and fully described by the parameters which were used to generate it from the model. The synthetic appearance and its vector representations are equivalent and inter-changeable. Visualising results is often convenient graphically, while logical operations are better thought of in terms of vectors.