No abstract

We present a Neural Network -Hidden Markov ModelHybrid for the recognition of cursive words which are represented as left-right sequences of graphemes. The proposed approach models words with ergodic HMMs and is designed for small vocabularies. A single neural network provides grapheme observation probabilities for all HMMs in order to compute the most likely word model. During the iterative EM like training of the hybrid, the HMMs provide the targets for the discriminant training of the neural network. An extension of the approach to letter models which can be concatenated in order to form word models and which allow for large vocabularies is also briefly discussed. We report results obtained on a large data base of words from French cheques, showing recognition rates close to 93% for the 30 word vocabulary relevant for French legal amounts.

No abstract

This paper presents the current state of the A2iA CheckReader TM -a commercial bank check recognition system. The system is designed to process the flow of payment documents associated with the check clearing process: checks themselves, deposit slips, money orders, cash tickets, etc. It processes document images and recognizes document amounts whatever their style and type -cursive, hand-or machine printed -expressed as numerals or as phrases. The system is adapted to read payment documents issued in different English-or Frenchspeaking countries. It is currently in use at more than 100 large sites in five countries and processes daily over 10 million documents. The average read rate at the document level varies from 65 to 85% with a misread rate corresponding to that of a human operator (1%).

The paper presents new A2iA bank check recognition systems designed to process handwritten and/or printed checks issued in France, UK or USA. All the systems have identical architecture and design principles, however, each of them contains a country-specific part and is trained with a country-specific data. Each system performs location, extraction, segmentation and recognition of courtesy-and legal amounts in a document image; as well as decision making to accept or reject the check. Recognition rate achieves 80-90%. In production mode, check acceptance rate is 60-75%, with misread rate corresponding to that of a human operator (close to 1%).