A model for assessing the reliability of document text field recognition

In this paper, we propose a model for assessing the reliability of simultaneous recognition for two text fields with the same content in a printed document. An example of such pairs could be the fields “amount” and “amount in words”.
The model analytically assesses the probability of independent recognition results in several fields to be coherent, while the fields in reality may be coherent and not coherent. We suggest a method for evaluating a single character recognition reliability that allows for a given multicharacter word recognition reliability threshold.