|
Abstract:
|
In image analysis, it is often desirable to reconstruct the boundary of an object in a noisy image. For example in mammography, boundary outline of the breast tissue is required as a part of a routine image assessment procedure to determine the risk of developing breast cancer. Various methods for boundary detection have been proposed (Markov random field models, active contour models, and stochastic models with polygonal shapes). This work introduces a novel method based on B-spline curves. The method allows to estimate a variety of complex contours, including boundaries of nonconvex sets and nonsmooth domains. The method is Bayesian and uses Markov chain Monte Carlo simulations to draw inference about the boundary. The method is illustrated on the simulated data and applied to recover a skin line on a digitized analog mammogram image. We also discuss the selection of a loss function.
|
