Improving semantic segmentation with graph-based structural knowledge - Archive ouverte HAL Access content directly
Conference Papers Year : 2022

Improving semantic segmentation with graph-based structural knowledge

Jérémy Chopin
• Function : Author
• PersonId : 1073969
Jean-Baptiste Fasquel
Harold Mouchère
Rozenn Dahyot
• Function : Author
Isabelle Bloch

Abstract

Deep learning based pipelines for semantic segmentation often ignore structural information available on annotated images used for training. We propose a novel post-processing module enforcing structural knowledge about the objects of interest to improve segmentation results provided by deep learning. This module corresponds to a many-to-one-or-none'' inexact graph matching approach, and is formulated as a quadratic assignment problem. Using two standard measures for evaluation, we show experimentally that our pipeline for segmentation of 3D MRI data of the brain outperforms the baseline CNN (U-Net) used alone. In addition, our approach is shown to be resilient to small training datasets that often limit the performance of deep learning.

Dates and versions

hal-03633029 , version 1 (02-01-2023)

Identifiers

• HAL Id : hal-03633029 , version 1
• DOI :

Cite

Jérémy Chopin, Jean-Baptiste Fasquel, Harold Mouchère, Rozenn Dahyot, Isabelle Bloch. Improving semantic segmentation with graph-based structural knowledge. ICPRAI 2022 - Third International Conference on Pattern Recognition and Artificial Intelligence, Jun 2022, Paris, France. pp.173-184, ⟨10.1007/978-3-031-09037-0_15⟩. ⟨hal-03633029⟩

102 View