Recognition of handwritten dates on bank checks using an HMM approach

Abstract

The article presents the first results of our system applied to the automatic recognition of handwritten dates on Brazilian bank checks. Considering the omni-writer context, we detail our recognition module dedicated to processing the month field. This module is based on the combination of holistic and analytical approaches with a fixed lexicon. Both approaches operate with a single explicit segmentation technique to provide a grapheme sequence for the purposed hidden Markov models of each recognizer. We show significant improvements when combining both modules to get a satisfactory recognition rate considering the small database images we work with. Finally, we present various perspectives for future work.