SoftMorph: Differentiable probabilistic morphological operators for image analysis

Lisa Guzzi

doi:10.1016/j.media.2026.104284

SoftMorph: Differentiable probabilistic morphological operators for image analysis

2026-09-02

0

OA

AI

DOI：10.1016/j.media.2026.104284

Original

Original request for help

Save

Abstract

En 中文

• SoftMorph makes morphological operations differentiable for deep learning models. • Converts any Boolean operator into differentiable probabilistic expressions. • Uses fuzzy logic and multilinear polynomials for smooth morphological filters. • Enables integration of morphology in CNN last layers and loss functions. • Improves topological accuracy in medical image segmentation on 2D and 3D datasets.

Keywords:

Morphological operations

Image analysis

Deep learning

Fuzzy logic

SoftMorph: Differentiable probabilistic morphological operators for image analysis

Abstract

Journal

Organization