Fuzzy and objectiveness integrated optimization of extended topological active net for multi object segmentation

Abstract

Image segmentation partitions an image to multiple objects. Topological Active Nets (TAN) and its extension Extended Topological Active Nets (ETAN) deforms meshes and composes them to fit to the objects to be segmented applying energy functional optimization. ETAN leads to local optima in cases of complex images with holes in it or with complex curves. In this work, an integrated fuzzy rule based learning and objectiveness measurement is used to optimize the ETAN. Fuzzy rule base is derived from training images for which segmented result is available as ground truths. Fuzzy rule base aids in decision for placement of links at segmentation boundaries. Objectiveness is foreground connectivity measure learnt with Laplacian Gaussian filter and used in decision for deletion of links in mesh for fitting complex shapes.