PDF version of this entire document
An NRR framework is constructed and explored for its behaviour to be understood. An important component of this proposed framework is the construction of an appearance model from NRR automatically, i.e. without manual intervention such as the placement of landmark points. This is shown in 1-D and also in 2-D at the later stages.
This chapter begins by providing an explanation of a simplified registration framework that brings one-dimensional image data into a state of alignment. This is followed by a more detailed explanation about the way such alignment is used to construct models of appearance directly.
Throughout this section, an approach is described for automatic construction of appearance models using a criterion related to MDL. However, in principle, one could replace this with other criteria. Technical details that are frequently described here refer back to earlier and similar work [,]. This work backed the idea of building shape models which are progressively refined by assessing their complexity.
This current work is distinct, but it is a cumulative step that owes credit to the inclusion of intensity data in the model. Finally, a method for evaluating such models will be described in a greater level of depth.