Scientific program

The Figure below summarizes the different levels targeted by the project :

  • The recognition task is separated into 4 levels, from T1 to T4.
  • For the tasks T1, T2 and T3, the cognitive approach is systematically compared to the state-of-the-art statistical approach on databases used for international handwriting recognition competitions (RIMES, IAM)
  • The task T1 aims at developing a model of the visual field integration based on a hierarchical stack of neural networks. The first neural networks that will be considered are self-organizing maps. See Topographica for a software implementation

  • The task T2 aims at using the open-bigram model proposed by Grainger & Whitney, 2004 for encoding the lexical information in a recognizer.
  • The task T3 aims at encoding the semantical proximity information of words with self-organizing maps in order to access to the semantical information in the recognizer.
scientific_program.txt · Last modified: 2010/11/21 21:42 by kermorvant