Publications by authors named "Adelie Garin"
Persistent Homology With Improved Locality Information for More Effective Delineation.
IEEE Trans Pattern Anal Mach Intell
August 2023
Article Synopsis
- Persistent Homology (PH) has been applied in training networks to identify curvilinear structures and enhance the quality of results, but current methods lack focus on the spatial location of topological features.
- This paper proposes a new filtration function that combines thresholding-based filtration from medical image segmentation with height function filtration used for shape comparison.
- Experimental results show that deep networks trained with this new PH-based loss function produce better reconstructions of road networks and neuronal processes than those trained with traditional PH-based loss functions.