Minimum Description Length (MDL) can be used to drive the correspondence selection process so that better models are gradually produced. It is used to form an objective function that guides a minimisation aiming to find a good similarity measure, i.e. minimum apparent difference amongst a set, thereby choosing good correspondences. It is yet unclear how it can be usefully applies for appearance models. Figure 5 which is shown later illustrates the contribution of MDL to the overall process. It inputs data that is jointly generated from the current model and some data set and it outputs an estimate that adjusts impending warps and affects the choice of numerous parameter values.
MDL was extensively used in  where for some given model, a message is passed which gets assigned a value of length that implies complexity. The message is encoded to encapsulate the relation between a data examples and the up-to-date appearance model. Sometimes an evaluation data example is based on the leave-one-out validation technique, meaning that a large number of examples will generate a statistical model and one will be used to evaluate the model4. If the model is too complex or not suitable to represent the data, the message that is passed will be greater in length.