1. General formalization of the problem

The problem of image registration involves estimating a geometric transformation or deformation of one image so that it is superimposed as closely as possible on another. This raises a number of questions: what information should be used to guide the registration process? How do you define the resemblance or distance between two images? What geometric transformation should be considered? How do you deform an image? How to estimate the best transformation? These questions lead to the following four elements characterizing a registration method:

• Attributes: these are the characteristics extracted from the images that guide the registration process. A distinction is made between extrinsic attributes (e.g. external markers attached to the subject) and intrinsic attributes (information extracted from the image, such as gray levels or extracted...