Boundary reconstruction in binary images using splines

In image analysis, it is often required to reconstruct the boundary of an object in a noisy image. This paper presents a new method, which relies on flexibility and computational simplicity of B-spline curves, to reconstruct a smooth connected boundary in a noisy binary image. Boundary inference is based on oriented distance functions yielding the estimator which is interpreted as a posterior expected boundary of the underlying random set. The performance of the method and its dependence on the image quality and model specification are studied on simulated data. The method is applied to reconstruct the skin-air boundary in digitised analogue mammogram images. •We present a new method for boundary reconstruction in noisy images using B-spline curves.•The method is Bayesian and uses oriented distance functions for inference.•The resulting estimator can be interpreted as a posterior expected boundary.•We study the performance of the method on simulated data.•We apply the method to reconstruct the skin-air boundary in mammogram images.