|
Abstract:
|
Given a collection of shapes from a population, one may wish to identify important points (landmarks) on these shapes which are preserved across the population. Landmarks are frequently used in statistical shape analysis, both in early and recent literature. The goal of this work is to provide an automated, model-based approach to detecting landmarks on shapes (represented as either open or closed curves). The model is based on a linear reconstruction of the collection of shapes, passing through the speci fied points. Inference is done in a Bayesian setting in order to efficiently estimate posterior distributions and quantify uncertainty in the landmark locations. The question of how many landmarks to select is also addressed in two different ways: a criteria-based approach, as well as incorporation into the Bayesian model (leading to a variable dimension parameter space). Efficient methods for posterior sampling are also discussed.
|
