Fast Lexical Post-Processing on Cursive Script Recognition

This paper presents a novel, fast approach to the problem of lexicon reduction. It is based on the concept of direct-addressing a table in which there is a slot of 1 bit for each component of 4 characters contained on the lexicon. In one of our experiments, the time needed for post-processing was reduced from 31sec to only 134ms, without any loss on recognition accuracy. The structure proposed is very flexible, and can be used as a front end for contextual post-processing on Handwritten or Handprinted Recognition. It is ideal to be implemented on parallel computers or hardware.