Soutenance de Leo Davy
| When |
Nov 13, 2025
from 02:00 to 04:00 |
|---|---|
| Where | Amphi A |
| Attendees |
Leo Davy |
| Add event to calendar |
vCal
iCal
|
This PhD work proposes a strategy for texture segmentation, that is, for analyzing images characterized by their statistical rather than geometric properties. The proposed approach relies on a multi-scale mathematical analysis (wavelet analysis), enabling the local and simultaneous estimation of regularity and anisotropy, two key features for describing textures observed in real-world applications.
The model is integrated within an optimization framework that combines the theory of iterative algorithms with the flexibility of neural networks. The method is evaluated on both synthetic and natural images, demonstrating strong robustness, even beyond idealized conditions. This explicit link with mathematically grounded methods ensures high performance while maintaining theoretical control over the learning process. Finally, a learning approach for neural networks applicable to a wide range of variational problems will also be presented and applied to various image restoration tasks.