Stochastic segmentation on images with uncertain data

Abstract

AbstractThe present work considers a stochastic segmentation method on images in the presence of noise within a PDE‐based image processing framework. Classical methods are not able to capture the error propagation of uncertain estimated input data and their impact on the final segmentation result, which can be of great importance for clinical decisions. Therefore, an intrusive generalized polynomial chaos (gPC) expansion for a stochastic level‐set based geodesic active contours method is proposed. Employing an operator splitting and a stochastic Galerkin projection a deterministic and symmetric non‐linear hyperbolic system can be obtained, which can be treated using common numerical methods.