LeRec: a NN/HMM hybrid for on-line handwriting recognition

  • Authors:
  • Yoshua Bengio;Yann LeCun;Craig Nohl;Chris Burges

  • Affiliations:
  • -;-;-;-

  • Venue:
  • Neural Computation
  • Year:
  • 1995

Quantified Score

Hi-index 0.00

Visualization

Abstract

We introduce a new approach for on-line recognition ofhandwritten words written in unconstrained mixed style. Thepreprocessor performs a word-level normalization by fitting a modelof the word structure using the EM algorithm. Words are then codedinto low resolution "annotated images" where each pixel containsinformation about trajectory direction and curvature. Therecognizer is a convolution network that can be spatiallyreplicated. From the network output, a hidden Markov model producesword scores. The entire system is globally trained to minimizeword-level errors.