Skip to Main content Skip to Navigation

Apprentissage d'un vocabulaire symbolique pour la détection d'objets dans une image

Abstract : We study the fondamental problem of selection of features describing a signal. This selection is done with respect to several signal processing task such image classification for instance.
We define a probability map on features to measure importance of these variables. We wish to learn a law that minimizes an efficiency criterion. We use then an exact and stochastic gradient descent algorithm to identify this law. We apply this approach to several problems like faces recognition, handwritten numbers classification, spam detection, toys data ... Then we study another approach that permits to make the features space evolve with time. This is done using a reflected diffusion, and a reflected jump diffusion algorithm. This algorithm is much harder to implement than the former gradient descent and we give a stochastic approximation to this reflected jump diffusion process. We apply this algorithm to the problem of face recognition.
Document type :
Complete list of metadata
Contributor : Sebastien Gadat Connect in order to contact the contributor
Submitted on : Thursday, March 3, 2005 - 11:11:29 AM
Last modification on : Wednesday, June 9, 2021 - 1:18:05 PM
Long-term archiving on: : Friday, September 14, 2012 - 11:50:09 AM


  • HAL Id : tel-00008642, version 1


Sébastien Gadat. Apprentissage d'un vocabulaire symbolique pour la détection d'objets dans une image. Mathématiques [math]. École normale supérieure de Cachan - ENS Cachan, 2004. Français. ⟨tel-00008642⟩



Record views


Files downloads