Prior shape models for boundary finding

Prior shape information has proven to be a key component of modeling for boundary finding when the target objects belong to a class of similar shapes. Prior shape provides specific constraining information needed in order to overcome noise, missing boundaries and confusing image information. While a number of different methods have been proposed for incorporating prior information, the most natural approaches use a Bayesian formulation where prior information and image derived information are combined by optimizing a posterior probability. Shape parameters derived from the statistical variation of the boundary in a training set can be used to model the object. Generic information such as from a smoothness constraint can be incorporated into the framework when additional flexibility is needed due to a small available training set.