Code recognition by means of improved distance and tuned dictionary

This paper presents a tuned Intelligent Code Recognition (ICR) model using a self-adapting dictionary, combined with a version of Levenshtein distance for checking word similarity. It is part of a broader project named PTAH (Procesamiento de Trámites con Asistente Hispanohablante), an intelligent chatterbot that uses this ICR to collect knowledge from documents. Due to its critical activity, it is mandatory to get the best result for the image to text conversion. The success in getting information determines the knowledge database quality and therefore the accuracy of the answers upon queries to the chatterbot. A blueprint of the global Project is included as well as the ICR, statistics progression obtained with the sequence of improvements applied to the ICR, and a final analysis of the failures.