Higher-Order Feature-Preserving Geometric Regularization

Abstract

We introduce two fourth-order regularization methods that remove geometric noise without destroying significant geometric features. These methods leverage ideas from image denoising and simplification of high contrast images in which piecewise affine functions are preserved up to infinitesimally small transition zones. We combine the regularization techniques with active contour models and apply them to segmentation of polygonal objects in aerial images. To avoid loss of features during the computation of the external driving forces we use total variation-based inverse scale-space techniques on the input data. Furthermore, we use the models for feature-preserving removal of geometric texture on surfaces.